You are on page 1of 59

‘Investigation towards the efficiency of a multi-cyclone

dust separator in biomass combustion’

Traineeship report

Author: Rob van Benthum


Student number: 0531335

Rapport number: WPC 2007.10

Traineeship supervisors: Ir. Carlo de Best (TU Eindhoven)


Dipl. –Ing. Josef Heinzle (Mawera)

Eindhoven, August 2007

Faculty of Mechanical Engineering


Division Thermo Fluids Engineering
Section Process Technology
Table of contents:
Preface . . . . . . . . 1

1. Introduction . . . . . . . 2

2. Cyclone theory . . . . . . . 3
2.1 Flow characterisation . . . . . . 3
2.2 Geometric coupling . . . . . . 4
2.3 Characteristic particle size . . . . . 6
2.4 Characteristic dust loading factor . . . . 8
2.5 Fractional efficiency . . . . . . 9
2.6 Separation efficiency . . . . . . 10
2.7 Pressure drop . . . . . . . 11
2.8 Model build-up . . . . . . 15

3. Measurements . . . . . . . 16
3.1 Introduction dust measurements . . . . 16
3.2 Setup and Equipment . . . . . . 17
3.3 Filter pre- and after-treatment . . . . . 20
3.4 Device comparison . . . . . . 21
3.5 Measurement plan and execution . . . . 22

4. Analysis . . . . . . . . 23
4.1 Model verification . . . . . . 23
4.2 Uncertainties in the model . . . . . 25
4.3 Uncertainties in the measurements . . . . 27
4.4 Accuracy analysis . . . . . . 27
4.5 Geometrical parameter determination . . . . 32
4.6 Cyclone design comparison . . . . . 34

5. Recommendations on the multi-cyclone geometry . 37

6. Summary . . . . . . . 39
6.1 Conclusions . . . . . . . 39
6.2 Additional remarks . . . . . . 40

7. References . . . . . . . 41

Appendices . . . . . . . 43
A.1 The considered (multi)-cyclone and its dimensions . . 45
A.2 Measurement Rapport . . . . . 46
A.3 Calculation of flue gas density . . . . 47
A.4 Normalized conditions . . . . . 48
A.5 Moisture content . . . . . . 50
A.6 Determination of the friction coefficient . . . 51
A.7 Particle size distribution . . . . . 52
A.8 Determination of the total cyclone wall-gas contact surface area 54
A.9 Swirl coefficient . . . . . . 55
Symbolic declaration:(rapport)
Ai Cyclone inlet surface [ m2 ]
Bi Cyclone inlet width [m]
CD Friction coefficient [-]
Ci Dust concentration [ kgm-3 ]
C in Dust concentration at multi cyclone entrance [ kgm-3 ]
C out Dust concentration at multi cyclone exit [ kgm-3 ]
d cr Critical exhaust flow diameter [m]
dp Particle diameter [m]
d p50% Characteristic particle diameter [m]
Dc Outer diameter of the cyclone separation chamber [m]
D0 Exhaust diameter of the cyclone separation chamber [m]
Fs Total contact area between the gas and the wall [ m2 ]
g Gravity acceleration [ ms-2 ]
Hi Cyclone inlet width [m]
kz Constant [-]
ki Axial expansion loss correction coefficient [-]
K Constant [-]
K0 Constant [-]
l Length of the cyclone outlet in the separation chamber [m]
L Length of the cyclone separation chamber [m]
mG Gaseous particle mass [ kg ]
mp Solid particle mass [ kg ]
m& in Mass flow rate at multi cyclone entrance [ kgs-1 ]
m& out Mass flow rate at multi cyclone exit [ kgs-1 ]
n Swirl exponent [-]
Δp … Pressure loss [ Pa ]
p dynamic Dynamic pressure [ Pa ]
p stat, after Static pressure after multi –cyclone [ Pa ]
p stat, front Static pressure in front of multi –cyclone [ Pa ]
Q Volumetric flow [ m3s-1 ]
r Trajectory radius [m]
re Characteristic radius [m]
ri Radius of entrance [m]
ur Radial particle velocity [ ms-1 ]
ut Tangential particle velocity [ ms-1 ]
v ax , 0 Averaged axial velocity in the cyclone exhaust [ ms-1 ]
vi Average cyclone entrance velocity [ ms-1 ]
vr Radial gas velocity [ ms-1 ]
vt Tangential gas velocity [ ms-1 ]
vt ,0 Tangential velocity at the position r = D o / 2 [ ms-1 ]
v~t ,c Dimensionless tangential velocity at position r = D c / 2 [-]
v TC Terminal velocity [ ms-1 ]
X geo Dimensionless geometric parameter [ -]
Cy 50 Rietema’s dimensionless parameter [-]
α Contraction coefficient for loss of angular momentum [-]
ζ… Pressure loss factor [-]
η Fractional efficiency [-]
η sep Separation efficiency [-]
θ Angle [ rad ]
λ Friction coefficient [-]
λw Wall friction coefficient [-]
μD Dust loading factor [ kgkg-1 ]
μG Dynamic viscosity of the gas [ Pa s ]
μ Gr Characteristic dust loading factor [ kgkg-1 ]
ρG Gas density [ kgm-3 ]
ρp Particle density [ kgm-3 ]
Γ sink Vortex strength of a sink component [-]
Γ vortex Vortex strength of a swirl component [-]
Preface:

Mawera is an innovative company in small- and middle-scale biomass combustion plants for
heat generation based in Austria with subsidiaries in Europe and Canada. They design and sell
complete installations from fuel storage and supply systems to chimneys and all that is in
between. Beside heat generation they are also involved in several projects for combined heat
and power generation from biomass. The traineeship execution took place at the home
company of Mawera in Hard am Bodensee in Austria, where approximately 150 employees are
stationed.

This report is about an investigation towards the efficiency of a multi cyclone dust separator.
This multi cyclone is used in a first cleaning step for burned gasses of biomass combustion
plants. It is meant to filter out small solid particles, down to approximately 1 μm, from the
burned gas stream right after the combustion chamber. The cyclone under investigation is
mostly adapted to biomass combustion plants up to 1 MW, running on non-contaminated
biomass fuels.

The main purpose of this report is the investigation of the separation efficiency of a given multi
cyclone, for different number of single-cyclones and for different kind of flow rates. There
should be found an optimum in separation, flow rate and pressure drop, for the current multi
cyclone. The results are supported by measurements on a test facility, available at the home
company of Mawera.

First there will be given a detailed theoretical background on particle separation efficiency,
starting with a force equilibrium method on a dust particle. This method is used to predict the
dp 50% ; the diameter of a particles that are separated with a 50 percent probability. This part is
followed by an investigation of the separation efficiency over the particle size and flow rate.
After a discussion on pressure loss determination, the theoretical part is closed with a short
introduction into the construction of a computer model.
Furthermore the test facility, a measuring plan, the measurement equipment are discussed.
In the third part the results of the measurements are compared to the prediction of the
theoretical values.
From the outcome of the comparison between the measurements and the theory, a fourth part
arises where some changes towards the design of the multi-cyclone are proposed.
This report is closed with a suggestion towards an optimal operating point and eventually an
improvement on the geometry of the multi-cyclone.

1
1. Introduction
The cyclone idea is first patented by Knickerbocker Co. Jackson, USA in 1886. The described
device is a funnel (reversed cone) with a top cover and a collection chamber at the bottom. The
top cover contains a cylinder in the centre functioning as an outlet. The device has a tangential
inlet at the top of the funnel. This device was designed for filtering woodchips from an
extractor flow in sawing processes. In the early thirties, it became clear and accepted that the
cyclone has more potential. From there the development broadened to separation from
thousands up to one micrometer [2].

As earlier mentioned a cyclone, see figure 1.2, is a particle separator, able to filter out small
solid and liquid particles from a gas or liquid stream. The operating range is dependent on the
cyclone geometry, flow rate and the density difference between the particles and the gas. The
main strength of the cyclone is the absence of moving mechanical parts.
The flow in a cyclone is
illustrated in figure 1.1. The
incoming gas stream is brought
to a swirling flow via a
tangential or spiral inlet or via a
swirl generator in the axial case.
This swirl causes centrifugal
forces. These forces are density
dependent and result in a move
of the heavy particles towards
the wall and the lighter gas
stream towards the inside of the
cyclone. Between the outlet figure 1.1: Schematic presentation of the flow in a cyclone. On
entrance and the bottom of the the right the trajectory of a particle and the gas stream are
cyclone the swirling flow is illustrated. On the left the development of the gas stream is
converted partly into a radial shown.
inward flow until the axial component turns. From this point a swirling flow of cleaned gas
goes up in the centre of the cyclone and leaves the cyclone through the outlet. The collected
particles at the wall are transported to the collection chamber via an axial flow at the wall,
eventually supported by gravity.
It can easily be seen that by increasing the flow, the separation efficiency goes up accompanied
by a larger pressure drop. In order to run at higher flow rates, more cyclones can be used in a
parallel setup [1],[12]. This is the main principle of a multi cyclone. See figure 1.2

2
2. Theory

2.1 Flow characterisation:

Cyclones are complicated devices for flow


calculations. The involvement of turbulence,
the interaction between particles and the
carrying gas, and the cyclone geometry makes
it hard to solve the equations of motion [2].
Even though, in a first approximation these
equations can be solved when some severe
simplifications and assumptions are carried out.
In a similar way as in [10] a start can be made Figure 2.1: Schematic presentation of the forces
at the equations of Newtonian motion of a on a particle in centrifugal separation
particle in cylindrical coordinates:

0.5 ⋅ C D ⋅ A ⋅ ρ G ⋅ (v r − u r )
2 2

+ (m g − m p ) ⋅ t
du u
mp ⋅ r =
dt mp r
(2.1)
0.5 ⋅ C D ⋅ A ⋅ ρ G ⋅ (vt − u t )
2
u ⋅u
+ (m g − m p ) ⋅ r t
du
mp ⋅ t =
dt mp r

The force change for both tangential and radial direction is a result of the change of the friction
force, the buoyancy force and the centrifugal force, see figure 2.1.
In expression (2.1) m p and m G are respectively the masses of the particle and the same particle
but then gaseous in [kg], d p presents the particle diameter in [m], v t and u t are the tangential
velocities of the gas and the particle in [ms-1], v r and u r are the radial velocities of the gas and
the particle in [ms-1], r is the trajectory radius of the particle in [m], ρ G is the gas density in
[kgm-3] and C D is a friction coefficient.
Under equilibrium condition, the particle moves in a circular orbit, meaning that its radial
velocity, u r , and its radial acceleration, du r dt are zero. Under the above condition the
tangential particle velocity is steady, so du t dt is zero. Substituting these assumptions into the
equations of motion, results in the following:

0.5 ⋅ C D ⋅ A ⋅ ρ G ⋅ v r
2
⎛ mg ⎞ u2
0= +⎜ − 1⎟ ⋅ t
mp ⎜m ⎟ r
⎝ p ⎠ (2.2)
0.5 ⋅ C D ⋅ A ⋅ ρ G ⋅ (vt − u t )
2

0=
mp

So the particles have the same tangential velocity as the gaseous phase: u t = vt .
Because the particle is in a steady orbit, and its radial velocity is zero, the radial velocity of the
gas is per definition the terminal velocity, v TC , of the particle, since the terminal velocity
represents the velocity of the particle seen from the gas.

3
If we now suppose that the radial inward gas flow is laminar and the particles are perfectly
spherical, a so called Stokes flow can be assumed around the particle and the friction
coefficient can then be presented as:

24 ρ G ⋅ d p ⋅ vTC
CD = Re = (2.3)(2.4)
Re μG

This expression for the drag coefficient holds for Reynolds numbers below one. Within the
above expression the Reynolds number is given as equation (2.4). Herein presents μ G the
dynamic viscosity of the gas in [Pa s] and v TC the terminal velocity in ms-1.
Combining equation (2.2) with (2.3) and (2.4) and replacing the masses by densities and
particle size leads to a relation between the particle size, particle position and particle motion:

(ρ p − ρ G )d p vt
2 2

vTC = (2.5)
18μ G ⋅ r

In this expression ρ p is the solid particle density in [kgm-3]. This expression can now be used
for a cyclone model.

2.2 Geometric coupling:


The working principle of a cyclone is based
on its swirling flow inside, which causes
centrifugal forces. According to ´Bart van
Esch´,[1] and ´Cristóbal Cortés´,[11], the
tangential component of the swirling flow
can in a first approximation described in
two parts:
In the outer part of the cyclone the
tangential velocity distribution is given by a
(reduced) free vortex ( vt ⋅ r = constant)
whereas in the core flow a solid body
rotation ( vt r = constant) exists. The
tangential velocity distribution is also
shown in figure 2.2.

As the particle concentration is sufficient


low (no flow disturbance by particles) the
particles can be assumed to behave in the
same way as the gas flow. Furthermore it is Figure 2.2: Presentation of the velocity distributions in
assumed that there exists a particle with a a cyclone. Right: the swirling flow. Left: 1: The radial
diameter d p that has a trajectory on a steady speed. 2: axial velocity distribution. 3: tangential
velocity distribution. ( Source: [1] )
radius r where it is theoretically neither
separated nor let trough. Than potential flow theory can be used to express the tangential and
radial gas velocity on the outside of the in terms of cyclone geometry.
In combination with equation (2.5) the particle size versus the particles steady radius can be
predicted.

4
The stream function for the flow in the outer part of the cyclone can be described as:

Γsin k Γ
Ψ (r , θ ) = − θ − vortex ln(r ) (2.6)
2π 2π

Herein Ψ (r ,θ ) represents the stream


function, Γsin k and Γvortex give the strengths
of the sink and the vortex component, and θ
is an angle in [rad]. The resulting velocity
expressions are depicted below:

1 ∂Ψ Γ
vr = = − sin k
r ∂θ 2π ⋅ r
(2.7)
∂Ψ Γvortex
vt = − ≈
∂r 2π ⋅ r

The unknown sink and vortex strengths can


be derived from the cyclone geometry and
the volumetric flow Q in [m3s-1], using a
specific control area in radial and tangential
direction:

Figure 2.3: Schematic representation of the cyclone Q = vt ( )⋅ B ⋅ H


DC
2 i i
(2.8)
( )
and its important dimensions. Do
Q = −2π ⋅ ⋅ (L − l ) ⋅ v r DO
2
2

In equations (2.8) D c presents the outer cyclone diameter in [m], D o the cyclone exit diameter
in [m], L and l the total length of the separation chamber respectively the length of the exit
duct in the separation chamber in [m] and B i and H i the inlet width and height in [m]. See also
figure 2.3.
In this way the vortex and sink strengths can be approximated and filled together with velocity
expressions (2.7) into (2.5), when off coarse v r is used for the terminal velocity. The result is a
relation between steady orbit particle position and the particle size:

36 μ G r 2 (Bi ⋅ H i )
2

dp = (2.9)
π (ρ p − ρ G ) ⋅ Q ⋅ (L − l )Dc2

5
2.3 Characteristic particle size:
A major parameter for cyclones is the particle size of a particle which is separated with a 50
percent chance on a critical radius r e . This characteristic particle size is called the d p50% .

The main problem now is the definition of the value of the characteristic radius r e . At this
radius the axial velocity is assumed to be zero, as can be seen from figure 1.2. Within this
radius, the particles are carried out of the cyclone. Particles outside this radius are carried to the
wall, and transported to the downside of the cyclone due to the downwards axial velocity at the
wall. In the most simple assumption according ‘Bart van Esch’ [1], re = D0 2 .

9 μ G Do2 (Bi ⋅ H i )
2

d p ,50% = [1] (2.10)


π (ρ p − ρ G ) ⋅ Q ⋅ (L − l )Dc2

In additional literature also other expressions can be found that are a variation on the derivation
of (2.10). The following expression can be found in the book from ‘VDI Verlag’ [2]:

μ G (Bi ⋅ H i )2
2n
⎛ D0 ⎞
d p 50% ⎜
= 9⋅⎜ ⎟ [2] (2.11)
⎝ 2ri ⎠ π (ρ p − ρ G ) ⋅ Q ⋅ (L − l )

In this expression the proportionality between the cone and the tangential velocity is corrected
with a swirl exponent n, which should take into account a correction for the ‘reduced’ free
vortex. The value of n can vary approximately between 0.4 and 0.8 [11]. In this expression, r i
represents the radius of the cyclone entrance in [m]. More exact, empirical relations of the
swirl coefficient can be found in additional literature [7],[11]. These relations are shortly
explained in appendix A.9.

A third expression can be found in the book of ‘Prof. Dr.-Ing Wolfgang Fritz’ [3]:

π v ax , 0 9μ G
d p 50% = D02 ⋅ ⋅ [3] (2.12)
4 vt , 0 π (ρ p − ρ G ) ⋅ Q ⋅ (L − l )

Herein v ax , 0 represent the averaged axial velocity in the exhaust duct of the cyclone in [ms-1]
and vt , 0 the tangential velocity in the exhaust duct of the cyclone at radius D o /2 in [ms-1]. An
often used relation between both velocities is derived from the angular momentum balance.
This results in:

vt , 0 D0
= (2.13)
v ax , 0 2αAi
+ 2λ ( L − l )
π ⋅ ri

λ represents the dimensionless friction coefficient and is determined in appendix A.3. α


Presents a dimensionless contraction coefficient for the loss of angular momentum due to the

6
cyclone inlet geometry and μ D represents the dust loading factor in [kgm-3]. The latter two are
expressed as:

m& D 2 ⋅ vi ⋅ ri
μD = α= (2.14) (2.15)
ρG ⋅ Q Dc ⋅ v t , c

In equations (2.14) m& D represents the mass flow rate of dust at the cyclone entrance in [kgs-1].
According [3], for a spiral cyclone inlet the dimensionless contraction coefficient from (2.15)
can be approximated as:

ri
α = 1+ 3 ⋅π ⋅ λ (2.16)
Ai

Equation (2.13) gives just an approximate value for the relation between the tangential and
axial velocity in the exhaust duct, independent of the flow rate. It is quite reasonable to assume
that this ratio differs for higher flow rates. Therefore, an empirical relation could be used for
the tangential velocity in the separation chamber. Such an empirical relation is given in section
2.6, equation (2.31).

Figure 2.4 compares the three expressions for particles, with a particle density of 800 [kgm-3].
As can be seen from figure 2.4, the three expressions give quite the same results:

18
Eq. (2.10)
Eq. (2.11)
16 Eq. (2.12)

14

12
dp50% in [um]

10

2
0 50 100 150 200 250 300 350 400 450 500
3
Flow in [m /h]

Figure 2.4: Comparison of the three expressions for the dp50%. Herein particles in biomass exhaust gas are
assumed. The particle density is put equal to 800 [kgm-3]. The biomass exhaust gas is assumed to be from
combusted wood with a moisture content of 30 mass% (d.b.) at a temperature of 175 degrees Celsius and an
absolute pressure of 857 mbar.

7
2.4 Characteristic dust load factor:
The dust load in the gas stream is a very important parameter in the separation process. This
dust load is normally given in the weight of dust per normal cubic meter of gas.
According ‘Prof. Dr.-Ing Wolfgang Fritz’ [3] and ‘E. Weber’ [5] it is proven in earlier research
that above a certain load factor µ D spiral dust flows build up on the wall right after the cyclone
entrance. This means that part of the dust is already separated at the entrance at these high dust
loads. In practice these spirals have the advantageous effect of turbulence damping and flow
stabilisation, resulting in a reduced pressure drop and better separation efficiency.
This result can be explained by taking a closer look to the flow around a particle: When a
particle enters a specific flow area, the total flow area is slightly decreased. This results in an
increase of flow velocity directly around the particle, accompanied with a decrease in pressure.
This lower pressure vanishes in a tail behind the particle. If another particle is close enough
behind the first particle, it will be sucked into the lower pressure tail of the first one, meaning
that the second particle follows roughly the trajectory of the first particle. So if the dust load is
sufficiently high, the solid particles are close enough to follow each other.

This dust load is approximated by a characteristic dust load factor µ Gr , expressed as:

1 3

λ ⎛⎜ 4 Ain Dc ⎞2 ⎛ vt , 0 Dc ⎞ 2 ⎛ 2(L − l ) ⎞
μ Gr = ⋅ ⎜α ⋅ ⎟ ⋅ ⎜⎜ ⎟ ⎜ ⎟
10 ⎝ π ⋅ Do 2 2ri ⎟ v D ⎟ ⎜ D − D ⎟ ⋅ 0.578 [3] (2.17)
⎠ ⎝ ax , 0 o ⎠ ⎝ c o ⎠

Equation (2.17) gives the characteristic dust load factor in [kgm-3]. The last number in this
expression counts for the correction of the particle with a 50 percent separation chance. This is
because this actual characteristic particle size is in most cases somewhat larger than the
calculated one. Another expression for this characteristic parameter can be found in the book of
‘E. Weber’[5]:
2
kz ⋅ λ Dc πDo 2 2π (L − l ) ⎛⎜ vt , 0 4Q ⎞

μ Gr = ⋅ ⋅ ⋅ ⋅ ⋅ ⎟
[5] (2.18)
⎛ Dc ⎞
3
2 Do 4 Ai Ai ⋅ g ⎜⎝ v ax , 0 πDo 2 ⎠
π ⋅ ⎜⎜ α ⎟⎟
⎝ 2ri ⎠

Herein g presents the gravity acceleration in [ms-2] and k z is a constant, which is approximately
2 ⋅ 10 −3 for average particle sizes of 10 micrometer. These two very complicated equations give
only a rough approximation of this characteristic dust load. This difficulty of approximation is
caused by the influence of many parameters on which the phenomena of dust-spiral-building
depends, such as the different velocities and their fluctuations and relations to each other, the
cyclone geometry, the wall friction, the particle distribution, etc. In most cases these equations
are used to estimate the order of the characteristic value. For the cyclone under investigation
the minimum value for which this phenomena can occur is about 0.02 [kgm-3]. This is roughly
a factor ten above the average dust load of primary exhaust gases from biomass combustion
chambers.

8
2.5 Fractional efficiency:
Fractional efficiency is very simply the separation chance for a certain particle size. Fractional
efficiency is hard to calculate, since it depends on many parameters such as cyclone geometry,
the flow characteristics and turbulence, dust load, particle distribution over the cyclone, etc.

In theory a step function at the d p,50% -point could be the most simple model for the fractional
efficiency. An expression of section 2.3 can then be used which includes the major geometrical
and flow characteristics. Unfortunately this idea is of limited use because the split between
separation and bypass is not fixed at a certain particle size but at a broad particle size area in
which the separation chance rises from 0 to 100 percent. Furthermore the separation chance
curve might not be symmetrical meaning that the dp 50% -point is not in the center of this particle
size area.
Although if it is assumed that the concentration distribution over the cyclone separation
chamber is homogeneous, a better approximation of fractional efficiency can be made using a
comparison between a total and a fractional volume. The fractional volume is then defined by
the radius of steady orbit for a certain particle size.
The steady orbit radius of a particle defines the boarder between the position of a particle that
is carried to the wall (separated) and the position of a particle with the same size that is carried
to the core flow. The number off particles with the same size that are outside a cylindrical
volume spanned by their steady orbit, divided by the number of particles with the same size in
the total cyclone separation chamber volume gives an efficiency for a certain particle size. For
a certain particle size range, this results in a fractional efficiency. An example of this method is
given in figure 2.5 for Mawera’s cyclone.
The main drawback of this approach is that a shortcut flow directly around the entrance of the
exit duct is neglected. This shortcut flow is much stronger than the general radial flow in the
cyclone separation chamber itself, resulting in a small amount of particles (of every size) that
are directly carried out of the cyclone. This is why the resulting fractional efficiency goes steep
to one. In reality the after the 50 percent fractional efficiency, the curve should become less
steep because of this effect.
1
Most expressions for particle `Bradley`
volume-comparison
efficiency are however derived on an 0.9 step-function @ d
p,50%

empirical basis using experimental 0.8


results and are related to the
0.7
characteristic particle size. In ´Erik
van Kemenade´ [12], such a relation 0.6
efficiency

is given according the method of 0.5


Bradley:
0.4

⎧ ⎛ d ⎞ ⎫⎪
X 0.3
⎪ ⎜ p
η = 1 − exp⎨− ⎜ − b⎟ ⎬ 0.2
d ⎟
⎪⎩ ⎝ p , 50 % ⎠ ⎪⎭
(2.19) 0.1

log(− log(0.5))
X = 0

log(1 − b )
0 5 10 15
particle size [um]

Bradley used for his S-curve: Figure 2.5: Bradley’s fractional efficiency (S-curve) given for
3 -1
b = 0.115 and X = 3. The S-curve for a volumetric flow of 200 m h . The step function (black line)
gives the d p,50% -point.
these values is given in figure 2.5.

9
2.6 Separation efficiency:

The separation efficiency gives the rate between the amount of separated mass versus the
amount of ingoing mass. This efficiency is very situation dependent, since it involves the
particle size distribution in front of the cyclone.
This separation efficiency can best be expressed as:

m& − m& out


∑ C (d ) − ∑ C (d )
in p out p

η sep (Q ) = in
dP dP
= Q⋅
∑ C (d )
(2.20)
m& in in p
dP
Herein present C in and C out the concentrations in front and after the cyclone in [kg/m3].
This can be further simplified in combination with the fractional efficiency, since the fractional
efficiency is the link between the particle size distribution at the entrance and outlet of the
cyclone. This is graphically presented in figure 2.6:

Figure 2.6: Graphical presentation of the relation between the fractional efficiency and the PSD at the
cyclone entrance and outlet.

The resulting expression is now given below:

∑C (d ) − ∑ (η (d , Q)⋅ C (d ))
in p p in p

η sep (Q) = Q ⋅dP dP


(2.21)
∑C (d )
dP
in p

So to predict the cyclone separation efficiency, one needs to know both the particle density
distribution in front of the cyclone and the fractional efficiency curve.
This separation efficiency is also the main goal of the measurements, discussed in the next
chapter.

10
2.7 Pressure loss:

Besides the characteristic particle size and separation efficiency of a specific cyclone, also the
pressure drop is important. This is because a cyclone is simply and solely driven by the gas
flow itself. In general the volumetric flow and the pressure drop to be conquered are coupled
via the kinetic (and or potential) energy content of the flow. An increased pressure drop results
in a decreasing volumetric flow, which then on its turn results in a worsening particle and
separation efficiency.
In the case of the biomass combustion plant design, an external fan is used after the cyclone to
conquer the pressure losses in the exhaust system. The operating conditions and the maximum
capacity and efficiency are thus dependent on the capacity of the fan.

In a cyclone, the total pressure loss can be prescribed to three general sources [5]:

1) Pressure losses in the inlet region of the cyclone. These pressure losses occur because
of strong directional changes in velocity and due to changing inlet cross sections.
2) Pressure loss in the separation chamber of the cyclone due to wall friction and friction
in the flow.
3) Pressure losses in the cyclone core and in the exhaust duct of the cyclone, due to strong
velocity changes in magnitude and direction.

In ‘VDI Verlag’ [2], ‘Prof. Dr.-Ing Wolfgang Fritz’ [3] and ‘E. Weber’ [5], an approach
according “W. Barth” and “E. Muschelknautz” is used to predict the pressure drop in a pure
gas driven cyclone. This method contains above three fractions of pressure loss separately.
These three pressure losses can be determined via loss factors. For the total pressure drop the
following equation can be set up:

Δp = Δp sp + Δp c + Δp o = ζ ⋅ ρ g ⋅ v ax , 0 = (ζ sp + ζ c + ζ o ) ⋅ ρ g ⋅ v ax , 0
1 1
( 2.22 )
2 2

Herein Δp sp is the pressure loss due to the spiral inlet geometry, Δp c is the pressure loss in the
separation chamber and Δp o describes the pressure loss caused by the exhaust duct. All
pressure losses are in [Pa].These loss factors are given as:

⎡ ⎤
⎢ 2 ⎥
⎢ π Do π ⎥ ⎛ 1 ⎞ ( 2.23 )
ζ sp =⎢ + ⎥ ⋅ ⎜1 − 2 ⎟
4 Ai D 4 Ai ⎝ α ⎠
⎢ 2 c ⎥
⎢⎣ Do π Do ⎥⎦
2

⎡ ⎤
⎢ ⎥ 2
Do ⎢ 1 ⎥ ⎛ vt ,0 ⎞
ζc = ⎢ − 1⎥ ⋅ ⎜⎜ ⎟ ( 2.24 )
2 ⎟
Dc ⎢ ⎛

vt ,0 2(L − l ) ⎞⎟ ⎥ ⎝ v ax , 0 ⎠
⎢⎜ 1 − ⋅ λ ⋅ ⎥
⎣⎢ ⎝ v ax , 0 Do ⎟⎠ ⎥⎦

11
4
2
⎛v ⎞ 3 ⎛ vt ,0 ⎞ ⎛ ⎛ vt ,0 ⎞⎞
ζ o = K ⋅ ⎜⎜ t , 0 ⎟ +⎜
⎟ ⎜v
⎟ + K0
⎟ ⋅ ⎜1 − ⎜⎜ ⎟⎟ ( 2.25 )
⎜ ⎟⎟
⎝ v ax , 0 ⎠ ⎝ ax , 0 ⎠ ⎝ ⎝ v ax , 0 ⎠⎠

In expression (2.25) the constants K and K 0 are equal to 4.4 respectively 2.0 for a sharp exhaust
duct edge and 3.4 respectively 1.1 for a rounded exhaust duct edge. As can be seen in appendix
A.1, the values for a sharp edge are applied for the involved cyclone. If the relation between
the tangential and axial exhaust velocities is greater than one, the third term from equation
(2.25) is cancelled.
In the above expressions, equation (2.13) can be used for the relation between the tangential
velocity at the radius D o /2 and the average axial velocity in the exhaust. Another possibility is
to use an empirical relation such as equation (2.31)
In this approach reference is made to the axial velocity in the exhaust duct, because the exhaust
duct is the main and central component of the cyclone where major pressure losses occur. It
has to be noted that in [5], it is stated that the calculation of the loss factor for the exhaust duct
is determined in compliance with an amount of simplifications and assumptions. For this
reason above method of predicting pressure drop is not always representative. Therefore a
reference to experimental results for the loss factor is advised.

A recent released and more accurate calculation (two to three orders of magnitude) of the
pressure loss in a cyclone is given in a paper from ‘Jianyi Chen’ [6]. Herein the combination
between loss factor, average inlet velocity and pressure loss is made in a same way using four
loss factors: An expansion loss at the inlet causes a first loss factor for the inlet. A contraction
loss at the entrance of the outlet tube results in an additional loss factor at the outlet. Then there
is also a frictional loss in the separation chamber between the gas flow and the cyclone wall
which gives the third loss factor. Finally there is a dissipation loss of gas dynamic energy in the
outlet flow, resulting in the fourth loss factor.

At the inlet of the cyclone the flow will expand both radially and axially. This results in a local
pressure loss at the entrance. For a volute (spiral) inlet this loss is given as:
2
⎛ 2 Bi ⎞
ζ 1 = ⎜⎜1 − k i ⎟⎟ ( 2.26 )
⎝ Dc + 2 Bi − Do ⎠

This pressure loss is only dependent on the flow and the geometry of the inlet. In equation
(2.26), k i is a constant which corrects for the axial expansion loss. Its usual value is
approximately 0.3.
The second loss occurs at the entrance of the exhaust duct. This loss occurs because of an
abrupt reduction of the flow area when the gases enter the exhaust tube from the separation
chamber:

16 Ai
2 ⎛ ⎛D ⎞
2

ζ 2 = 4.5 ⋅ ⋅ ⎜1 − 3⎜⎜ o ⎟⎟ ⎟ ( 2.27 )
π 2 Dc 4 ⎜ ⎝ Dc ⎠ ⎟
⎝ ⎠

The third expression is the swirling loss. Because of the friction between the gas and the
cyclone wall due to gas viscosity, swirling energy is lost, appearing as an additional pressure
loss. Because of this loss, the tangential flow in the cyclone is a combination of a quasi-free

12
and quasi-forced vortex. This can be expressed as an exponent in the free vortex relation: this
exponent is called, as seen earlier, the swirl coefficient, and is explained in appendix A.6. It
can be seen that this third loss factor is dependent on the tangential flow at the wall, the wall
contact surface area and the friction coefficient:
1.5 n
λ ⋅F ⎛D ⎞
ζ 3 = w s ⋅ v~t ,c 3 ⋅ ⎜⎜ c ⎟⎟ ( 2.28 )
0.9 ⋅ Ai ⎝ Do ⎠

Herein F s is the total contact area between the gas and the wall in [m2]. A calculation of this
surface area can be fount in appendix A.5. v~t ,c is the non dimensional tangential velocity
(v t ,c vi ) at the wall, n is the swirl exponent and λ w is the wall friction coefficient, determined
in appendix A.4.
The last loss factor considers the dissipation loss of the gas dynamic energy in the outlet flow.
−n
⎛d D ⎞ 16 Ai
2

ζ 4 = v~t ,c 2 ⋅ ⎜⎜ cr o ⎟⎟ + ( 2.29 )
⎝ Dc Dc ⎠ (
π Do − d cr
2 2
)
2 2

This is a complicated factor and it has the largest contribution to the total pressure loss. The
determination of this loss is done using the fact that the axial flow is negligible in the vicinity
of the core and is quite high in the outer region of the exhaust duct. Based on the axial flow
profile, a quasi free vortex in the annular region and a quasi forced vortex in the core region of
the exhaust duct, a critical diameter d cr can be determined, separating the two flow fields. This
critical exhaust flow diameter, d cr expressed in [m], is used to determine the dissipation loss.
Only the tangential velocity distribution and the critical exhaust flow radius are unknown, but
can be determined according [7], using the correlations:

⎛ ⎛D ⎞ ⎛D ⎞
2

d cr = Dc ⋅ ⎜ 0.38⎜⎜ o ⎟⎟ + 0.5⎜⎜ o ⎟⎟ ⎟ ( 2.30 )
⎜ ⎝ Dc ⎠ ⎝ Dc ⎠ ⎟
⎝ ⎠

−0.21 0.16
⎛ π Dc 2 ⎞ ⎛D ⎞
1.11 ⋅ ⎜⎜ ⎟
⎟ ⋅ ⎜⎜ o ⎟⎟ ⋅ Re 0.06
vt ,c ⎝ 4 A ⎠ ⎝ Dc ⎠ 4 ρ g Ai vi
v~t ,c =
i
= Re = ( 2.31 )
vi 4 ⎛ π Dc 2 ⎞⎛ Do ⎞ πDo μ g
1+ λ ⋅ ⋅ Fs ⋅ ⎜⎜ ⎟⎜
⎟⎜ D ⎟

πDc 2 ⎝ 4 A i ⎠⎝ c ⎠

In expression (2.31) F s presents the total gas-wall contact surface in m2. The total pressure
drop is now given as:

1
Δp = (ζ 1 + ζ 2 + ζ 3 + ζ 4 ) ⋅ ρ g ⋅ vi ( 2.32 )
2

Herein represents vi the average cyclone entrance velocity in [ms-1].

13
In above pressure drop calculation it is also possible to take in account the dust loading of the
gas. Therefore, the density ρ g needs to be replaced with (ρ g + C i ) in equations (2.26) and
(2.32) and with (ρ g + {1 − η sep }C i ) in expressions (2.27) to (2.29) and (2.31) and (2.32). Herein
C i presents the dust concentration in [kgm-3]. If the cyclone separation efficiency η sep is
sufficiently high (suppose in the order of 90 to 100 percent), then the separation efficiency term
term can be neglected and (ρ g + {1 − η sep }C i ) changes in ρ g.
Not only is the density influenced by dust-loading, but also the wall friction. Therefore λ w
needs to be replaced by λ where the latter is determined in appendix A.4. The last concern is
the tangential velocity at the wall. A new correlation or correction needs to be used in dust
laden situations [7]:

v~t ,c
v~t′,c = 0.27
( 2.33 )
⎛C ⎞
1 + 0.35⎜ i ⎟
⎝ ρ g ⎠

Using these equations, the following results can be viewed for the single cyclone where
biomass combustion flue gas is used at a temperature of 175 degrees Celsius:

50
Chen & Chi
45 Barth & Muschelknautz

40

35
pressure drop in [mbar]

30

25

20

15

10

0
0 100 200 300 400 500 600
3
Flow in [m /h]

figure 2.5: The pressure drop over the single cyclone for biomass combustion flue gas at a temperature of
175 degrees Celsius. The red (upper two) lines gives the method of ‘Barth and Muschelknautz’ according
[2],[3] and [5]. The blue (lower two) lines gives the method of ‘Chen and Chi’ according [7]. The continuous
lines are for the pure gas laden situation and the dashed lines are for a dust laden situation (water droplets)
with a loading factor of 0.001 kg particles per kg flue gas. As in accordance with [7], the pressure drop
decreases for small dust loadings.

14
2.8 Model build-up:

In order to make a prediction of the separation efficiency of the multi cyclone under
investigation, a numerical computer model of a single cyclone is set up in a mathematical
environment, called Matlab.
After parameter specification, a prediction is done of the characteristic particle size over the
volumetric flow in the same way as described in section 2.3.
In combination with the characteristic particle size a fractional efficiency is introduced for each
flow rate, as described in equation 2.17 in section 2.5. This fractional efficiency is then
combined with the assumed particle size distribution in front of the cyclone, introduced in
appendix A.4. The separation efficiency for each flow rate is then equated according equation
(2.19) in section 2.6.

The second part of the model concerns the prediction of the pressure drop over a single
cyclone. The pressure drop is also calculated for different flow rates over the cyclone,
according the two methods in section 2.7.

The results of this model are to be compared with measurements on the test facility.

For the model a realistic cyclone operating condition of 175 degrees Celsius is taken with an
atmospheric pressure of 960 mBar. The density of the flue gas is determined as given in
appendix A.3. Herein an oxygen content of 13 Vol.% (related to dry flue gas) is assumed with
perfect combustion. The particle density is set to 800 kg/m3. This is somewhat less than the
usually used density of water since the particles are assumed to be porous.

15
3. Measurements:
In order to support and verify the cyclone performance model, measurements where performed
at the Mawera test facility . This test facility contains a multi-cyclone, just like the ones used
on commercial scale biomass combustion plants from Mawera.
The main purpose of these measurements is to determine the total fly ash concentration before
and after the cyclone and the pressure drop over the cyclone for different volumetric flow rates
over the cyclone at maximum performance (~350 kW) of the test facility.
In this chapter, first a global overview of an absolute dust measurement is given. Next the
combustion installation is shortly described, including the measurement setup. Thereafter the
pre- and after-preparation of the used filters is discussed, followed by the method of data
processing. This chapter ends with the results.

3.1 Introduction dust measurements:

The measurements, done at the Mawera test


facility, are absolute dust concentration
measurements. The measurements are done
according to the VDI guideline 2066, [9]. With
a heated probe exhaust gasses are iso-kinetically
extracted from the main flow for a period of at
least 30 minutes. Isokinetic extraction means
that the gas velocity in the probe inlet is equal
to the gas velocity in the main flow.
A changeable filter element is placed in the
probe. This filter collects all dust particles from
the flow trough the mouthpiece of the probe. It
is important to note that we measure a total
amount and not a distribution.
It can be assumed that this filter captures
particles down to approximately 1 µm. So only
the aerosols pass this filter. Since we are only
interested in the particle range that can be
separated with the multi-cyclone (coarse fly ash
particles), this is not a problem. figure 3.1: Dust measurement setup according
Via the weight of the filter element before and VDI 2066.
after the measurement, the dust mass is
determined. A combination of this dust mass with the time and the flow rate trough the probe,
gives then the absolute dust concentration.
This absolute dust concentration is formally recalculated for normal conditions, dry flow and
13% O 2 in the exhaust flow.

16
3.2 Setup and Equipment:

In order to investigate the separation efficiency of the multi cyclone, dust, pressure and
temperature are measured before and after the cyclone.
The pressure measurements are done with a Prandtl-type Pitot tube. The dynamic pressure is
then used to calculate the volumetric flow. The temperature and static pressure are used to
calculate the volumetric flow under normalized conditions. The temperature measurements are
done with Pt-100 thermo resistors.
The dust measurement behind the cyclone is executed with a probe like measurement device,
called a tubular filter device, described in VDI 2066, [9]. This device contains the filter bush,
and is inserted completely into the flow. See picture 3.1 and 3.2. The flue gas is extracted by an
external pump through the filter bush, some silica gel based filters and a gas meter. See also
figure 3.4. The gas meter measures the dry volumetric flow, the under pressure and gas
temperature. These values have to be read manually from the device. This is done at the
beginning, the middle and the end of each measurement. The normalized extracted volume can
be determined from these values. The timing is done manually for each measurement.

figure 3.2: The tubular filter device as mend in the VDI 2066 guideline. Left: a picture of the used device.
Right: a cross section of the device. The area spanned by d2 and l3 is filled with filter material.

The mouthpiece of the tubular filter device is tuned to the average velocity of the flue gas.
Since it is known that, in situ, the pump has a capacity of ~2 m3h-1 and the main flue gas
velocity is at maximum performance approximately 7.5 ms-1, a mouthpiece of 10 millimeter
diameter is used.
The dust measurement in front of the cyclone is done with a lance like device. The device, the
´STMG40`, is fabricated by `Afriso Euro Index´. The filter is here mounted into a heated
handle. The handle contains a ~50 cm long lance and a fixed mouthpiece of approximately 8
millimeter belonging to an inlet velocity of ~ 4 ms-1. This mouthpiece can not be changed. The
lance is inserted into the flow. The gas suction device is a controlled device which has a
regulated pump capacity of ~9L/h. Also here the gas is dried with silica gel based filters. This
device is shown in figure 3.3:

17
figure 3.3: The STMG 40 dust measurement device. Left: the lance and handle. The filter is inserted into
the white cylindrical part on the back of the handle. Right: The gas suction device is shown. The device is
regulated at a flow rate of ~9 L/h.

The STMG 40 is standard equipped with a measuring time of 15 minutes. For these
experiments 30 minutes are required. Therefore the measuring time is internally changed to 30
minutes. In order to validate the pump capacity, the flow rate of the device is measured with
the gas meter of the tubular filter device. Via the pressure and temperature on the gas meter,
normal conditions are derived. The results are depicted below in table 3.1:

tabel 3.1: Result of the flow rate validation experiments. The values are derived from eight measurements,.
In these measurements the pressure loss over the filter is simulated with a valve, so that in situ operation
conditions are approached.

Average volume (30 minutes) Maximum deviation Standard deviation

3
1.860 ⋅ 10 −2 nm3 10.283 ⋅ 10 −2 nm3
0.2546 nm ( 7.32 % ) ( 4.04 % )

figure 3.4 : Suction equipment of the tubular filter device. This equipment
contains the water filters, the gas meter and the pump.

18
In figure 3.5 an overview is given of the experimental setup. The measurement position in front
of the cyclone is between the boiler and the bend as can be seen from figure 3.4. The position
after the cyclone is just in front of the suction fan, after two bends in the exhaust duct.

figure 3.5: The schematic presentation of the total measurement setup as built in the test facility at Mawera.

As mentioned earlier, the dust concentrations are calculated at 13 vol.% oxygen in dry flue gas.
Therefore the volumetric flow trough the filter has to be corrected. This correction is given in
appendix A.3. In order to do this correction, the actual volumetric fraction of oxygen in dry
flue gas is measured with a gas analyzer (shown in figure 3.4). This fraction is stored via a
computer in a data file. The average oxygen content can be calculated from this data file for
each measurement.
In the same way temperature and static and dynamic pressure are stored into the data file. From
these values, the average the normalized flow rate of the flue gas can be calculated, just like the
average pressure drop over the multi-cyclone for the duration of the measurement.

19
3.3 Filter pre- and after-treatment:
In accordance with VDI guideline 2066, the filter element needs to be prepared before they are
used. A filter element of the tubular filter element consists of a metal bush, a small grating and
the filter material, which can be exchanged. A filter bush is shown in figure 3.6.
The filter material, quartz wool, is obtained from the company Mercateo (1).

The stepwise pre-measurement preparation of the filter bushes is stated below:

1) The filter bushes are numbered with a water proof pencil


2) The filter bush filled completely with quartz wool, while slightly pressed with the
thumbs
3) The quartz wool in the filter bush is saturated with tap water and pressed with the
thumbs into the bush until two third of the bush is filled with the filter material.
4) The filter bush is dried for at least 10 hours in a drying furnace at a temperature of 105
degrees Celsius.
5) The filter bush is mounted into the probe. Environmental air is sucked into the probe
trough the filter bush for at least 5 minutes to assure that broken most fibres are
removed from the filter bush.
6) The filter bush is again dried in a drying furnace for at least one hour at a temperature
of 105 degrees Celsius.
7) After drying the filter bush is cooled for at least 1.5 hours in a disiccator. This is a large
closed bowl filled with silica gel.

The filter bushes are taken out of the disiccator just one to two hours before the measurement
and straightaway weighted.

The pre-measurement preparation of the filter elements of the STMG 40 is somewhat different.
The filter element is a complete cartridge, obtained from: Bundesverband des
Schornsteinfegerhandwerks-Zentralinnungsverband (2).

1) A new filter cartridge is put into an aluminium canister. All used canisters are
numbered. The canister is closed with a corresponding aluminium lid during transport.
2) The cartridge taken out of the canister and is placed into the handle. Environmental air
is sucked through the filter for a period of at least 5 minutes in order to reduce the fibre
loss.
3) The filter is taken out of the device and is put into the canister again. The canister
without lid is then placed into the drying furnace (105 degrees Celsius) for at least one
hour.
4) The canister without lid is cooled into a dessicator for a period of at least 1.5 hours.

A canister with cartridge is taken out of the dessicator just one to two hours before the
measurement. They are straightaway weighted. The weight is determined from the canister
with cartridge and without lid.
After the measurement the filters have to be prepared again. This after-preparation concerns the
last two steps of the pre-preparation for both filter types. The filters are again weighted. The
weight difference gives the dust mass.
(1) Mercateo , www.mercateo.com , service@mercateo.com
(2) Bundesverband des Schornsteinfegerhandwerks-Zentralinnungsverband , Westerwaldstraße 6 , Sankt Augustin (D)

20
3.4 Device comparison:

Since the dust concentration measurements in front and after the cyclone are performed with
different kind of devices, both devices are compared with each other. This comparison is split
into two parts: The comparison of the sucked volume, and the comparison of the measured dust
concentration.

The comparison of the sucked volume is done by coupling the gas meter of the ´tubular filter
device´ to the STMG40. The results from this volume comparison are given earlier in table 3.1.

The comparison of the measured dust concentration is done by measuring the dust
concentration after the cyclone with the two devices simultaneously. The measurement setup
for device comparison is shown in figure 3.6. The distance between the two devices is
approximately 50 centimeter.
The results per measurement fluctuate quite a lot between the two devices. However, averaged
over the measurements are the results approximately equal. The results are depicted below in
table 3.2 and table 3.3:

Tabel 3.2: Average measured dust concentration and corresponding standard deviation, averaged per
device and per measurement. All values are related to 13 Vol.% oxygen and dry flue gas.

Tubular filter device STMG40


Average value 40,02 mg/nm3 41,30 mg/nm3
Standard deviation 12,17 mg/nm3 12,78 mg/nm3

Tabel 3.3: Average difference in concentration, and corresponding maximum


and standard deviation, between the two devices, averaged per measurement.
All values are related to 13 Vol.% oxygen and dry flue gas.

Average difference 9,4 mg/nm3


Standard deviation 2,56 mg/nm3
Maximum difference 14,1 mg/nm3

Figure 3.6: A picture of the device comparison measurement. The left device (blue handle) is the heated
probe from the gas analyzer. The distance between the STMG 40 (yellow handle) and the tubular filter
device (lance on right of picture) is approximately 50 cm.

21
3.5 Measurement plan and execution:

As mentioned earlier three important parameters are to be measured: the pressure drop,
volumetric flow rate and the total dust concentrations before and after the multi-cyclone.
The first two are measured with the Prandtl tubes, and the last one with the devices described
in section 3.2.
To achieve different flow rates over a single cyclone, the number of single cyclones in the
multi-cyclone is changed. This is done in such a way that three quite reasonable divided flow
rates are obtained.

The test facility is calculated at a maximum


flue gas flow rate of approximately 1400 m3
per hour. The multi-cyclone in the test facility
is originally equipped with 10 single
cyclones.
To keep the same setup of single cyclones in
the multi-cyclone, there is chosen to remove
two times three cyclones from the multi- Figure 3.7: Layout of single cyclones in the multi-
cyclone. The setup in the multi-cyclone and cyclone, for the three experiments. The depicted
the belonging flow rate per single cyclone is flow rate is the flow rate over a single cyclone.
given in figure 3.7:

It is difficult to determine the required number of measurements per number of cyclones,


because one does not know the fluctuation in the data on forehand. Just one day later, after
processing and evaluating all data, it becomes clear if more measurements are required.
Since the dust concentration measurements are the most error sensitive due to their manual
execution, the smallest number of demanded correct dust concentration measurements (in front
and after the cyclone simultaneously for determining separation efficiency) is set on forehand
to 7. This number proved to be sufficient for reliable and reproducible results.

To accept or reject a measurement, three criteria are handled:


First the operating power during a measurement must be in the range of full-power (~350 kW).
Second the dust concentrations must be of realistic order. From external data, originating from
earlier tests at the Mawera test facility, acquired via the University of Technology of Graz
(Austria),[13], it becomes clear that the dust concentration in front of the cyclone should be in
the order of 100 to 300 [mg/nm3, 13%O 2 ]. The concentration behind the cyclone is expected to
be somewhere in the range of 10 to 30 percent of the ingoing dust concentration according the
theoretical model.
Finally both dust concentration measurements have to be within their realistic range for
deriving separation efficiency.
If a measurement satisfies these three criteria, then it is called a properly executed
measurement and can be used for calculation separation efficiency.

After each measurement, data is processed, and all dust concentrations and flow rates are
normalized as explained in appendix A.3.

22
4. Analysis:
In this section the results from the measurements are to be compared with the theoretical
results from the model. The comparison is followed by a clarification of possible error sources
and uncertainties in the measurements and the model.
Furthermore, a cyclone specific geometrical parameter is introduced for comparison with other
cyclone designs. Finally the cost-efficiency for the current multi-cyclone is characterised by
comparing the pressure prop over the separation efficiency.

4.1 Model verification:

Below the results are presented from the model and the measurements. Figure 4.1 gives the
separation efficiency over the volumetric flow. Figure 4.2 gives the pressure drop over the
volumetric flow.
All the measurements were done with wood chips as fuel and with a moisture content of
approximately 30 Vol.%.(w.b.). Since the pressure and temperature fluctuate over time, all
volumetric parameters are recalculated to the model condition of 175 degrees Celsius and an
absolute pressure of 960 mBar. This is done to compare the measurements with the theoretical
results of the model.

100

95

90
Rendement in [%]

85

80
model (PSD from fig A.2)

measurements

75
mean per number of cyclones

model (PSD from fig 4.3)

70
0 50 100 150 200 250 300 350 400 450 500
3
Flow in [m /h]

Figure 4.1: Comparison of the measured separation efficiency, with the theoretical results from the model.
The measurements are done with 4, 7 and 10 single cyclones in the multi-cyclone. The triangles give the
mean point of the measurements for a certain number of single cyclones.

23
50
Chen & Chi

45 Barth & Muschelknautz

measurement data
40
Emperical relation eq.(4.1)

35
pressure drop in [mbar]

30

25

20

15

10

0
0 100 200 300 400 500 600
3
Flow in [m /h]

figure 4.2: Comparison of the calculated pressure loss with the measured pressure loss. The variation over
the flow rate is also here realized by the number of single cyclones in the multi-cyclone.

From figure 4.2 it can be seen that the pressure drop prediction is in accordance with the
experimental results.
Pressure losses for cyclones are often related to the inlet velocity with the use of a pressure
drop coefficient. (as explained in section 2.7). This pressure drop coefficient can empirically be
related to the inlet Reynolds number in a same way as described in ‘Erik van Kemenade´ [12].
For the cyclone under investigation the following expression is derived:

Q ρ g ⋅ Di
(Δp )static {
= 18.3469 ⋅ Re i
−0.0545
}ρ v
g i
2
; Re i > 500 ; Re i =
Ai

μ
(4.1)(4.2)

This relation is given in figure 4.2 with the black dotted line. Above expression gives the static
pressure loss, (Δp )static in [Pa].

Somewhat different holds for the measured separation efficiency. From figure 4.1 it can be
seen that the measured values are quite below the predicted curve. It is quite assumable that
this difference is caused by two very important assumptions in the model which are explained
below.

24
4.2 Uncertainties in the model:
As stated above, the main uncertainty comes probably from the two major assumptions in the
model:

1) The supposed particle size distribution in front of the cyclone as given in appendix A.6
2) The supposed shape of the fractional efficiency; the S-curve.

The exact particle size distribution on the measurement position in front of the cyclone is not
known, since no particle size measurements were done. Because a particle size distribution is
necessary to predict the separation efficiency, one is assumed on data from other measurements
(as explained in appendix A.6)
Particle size distributions in biomass combustion are very dependent on combustion unit
geometry, the kind of fuel and its composition, the moisture content, the degree of operating
power, furnace temperature, the amount of excess air, etc. Therefore it is obvious that the
assumed particle size distribution might not suit the actual particle size distribution in the test
facility of Mawera at the time of the measurements.

In the assumption of the particle


150
size distribution (appendix A.6),
aerosols are neglected since it
was assumed that aerosols
(particles smaller than 1 µm)
pass the filters from the dust
Concentration [ mg/nm ]

100
3

measurement devices anyway.


So the distribution below one
micron is not included in the
model. There is however a
possibility that the filters do 50

catch (some part of) the aerosol


peak, so the assumption has
certainly an influence on the
inaccuracy of the model.
0
From ´Aerosol Technology´,[15] 10 10 10
-3 -2
10
-1 0 1
10
2
10
3
10
it can be found that most particle Figure 4.3: The bimodal distribution
Aerodynamic Particle size [ um ]
of the particle size mass which
density distributions can best be is used in the model to show the effect of aerosols on the separation
described by a log-normal efficiency in figure 4.1.
distribution. In [14] a graphical
presentation of a complete bimodal particle size distribution for wood, chips, bark and waste
wood can be found. The distribution for wood chips can be approximated with the summation
of a log-normal distribution for aerosols and a lognormal distribution for coarse fly ash. The
corresponding values of the mean and standard deviation are given in table 4.1. From [13] it
can be seen that for wood chips, the fraction of total aerosol concentration is approximately 5
percent of the total coarse fly ash concentration. The resulting particle size distribution is
shown in figure 4.3.

25
Tabel 4.1: Values corresponding to the log-normal distribution as an approximation for the complete
particle size distribution of wood chips. 1) The standard deviation and the mean particle size in the log-
normal distribution are the logarithms of the values in the table. A further explanation of the log-normal
distribution is given in appendix A.6.
Aerosols Coarse fly ashes
1)
Mean particle size 0.015 µm 37 µm
Standard deviation 1) 2.75 µm 2.25 µm

The effect of expanding the PSD with the aerosol-peak, on the prediction of the separation
efficiency is shown with the black dotted line in figure 4.1. As can be seen, the influence of the
ignorance of the aerosol peak is quite small, so the ignorance of the aerosol peak is verified.
Therefore it might be expected that the distribution of the coarse fly ash peak was somewhat
broader during the measurements, which also result in a flatter separation efficiency curve.

The fractional
efficiency on the other
hand is a very cyclone-
geometry and flow rate
dependent function.
There are several
possibilities to
calculate this fractional
efficiency theoretically,
but, as stated earlier,
the methods are not
always representative.
Therefore usually
simple relations for the
S-curve are derived
from measurements,
which are specific per
cyclone geometry and
Figure 4.4: Effects of different fractional efficiencies on the separation
corresponding efficiency of a cyclone. On the right: the different fractional efficiencies with
operating conditions. their corresponding parameters b and X (see also equation 2.17). On the left:
For the cyclone under the predicted separation efficiency and the experimental values.
investigation this S-curve is unknown. Therefore a model is needed for this S-curve.
The separation efficiency model uses the expression of Bradley for an S-curve. However, the
parameters b and X in this expression are still unknown for the cyclone under investigation and
are certainly not the same has the ones from Bradley’s design. This brings again an uncertainty
in the model. Figure 4.4 shows the effect of different S-curve shapes on the separation
efficiency.
One might conclude from figure 4.4 that for the cyclone under investigation, a quite flat S-
curve (like the dashed green line in figure 4.4) is more appropriate.

A major improvement can be achieved on these two problems if the particle density
distributions in front and after the cyclone are measured (for example with an impactor) for
different flow rates over the cyclones. In this way one knows the exact particle size distribution
in front of the cyclone. The exact fractional efficiency can then be determined via the particle
size distribution after the cyclone. The d p50% -point can then be determined exactly and used for
comparison with other cyclone designs. This comparison will be explained in section 5.4.

26
Another less important but not negligible point which introduces an uncertainty is the
knowledge of some physical properties of biomass combustion gases and the particles in this
flue gas. For the particles in flue gas from wood chips combustion, a density of 800 kg/m3 is
assumed. The flue gas density is approximated on a basis of ideal combustion as given in
appendix A.3. The viscosity is assumed to be equal to normal air. This causes some deviations
from the real situation.

4.3 Uncertainties in the measurements:


The difference between the predicted and measured values for the separation efficiency might
not only be caused by uncertainties in the model, but also by faults and uncertainties in the
measurements.
A dust concentration measurement is a very complex manual measurement. First one needs to
prepare a filter for the measurements. After pre-preparation, the filters are weighted. After a
measurement, a filter is prepared again (as described in section 3.3) and then again weighted.
Every contact between the filters and the environment (like during the assembly and
disassembly of the filters in the devices) between the two weight determinations, can have
influence on the weight difference of the filter.
If a filter falls over in the drying furnace, the disiccator or on the balance, the measurement can
be assumed to be lost, because of the (possible) spoiled dust. These cases are however
excluded from the final data.
Then off course one needs a perfect, faultless, measurement in front and after the cyclone
simultaneously for a determination of the separation efficiency. If not, a measurement is
assumed to be inappropriate for separation efficiency determination.
For calculation of the normalised vacuumed volume over the filter during a dust concentration
measurement, one needs to read out the gas temperature and the under pressure of the gas
manually. This is done three times during the measurement. So the average value might not
correspond with the real average. Then off course one can make a mistake during manual read
out of the vacuumed volume, temperature and pressure.
Other possible deviation sources are for example the position of the Prandtl tube compared to
the centre of the flow, the accuracy of the Pt-100 temperature sensors and the calibration offset
between sensors of the same type.

4.4 Accuracy analysis:


The uncertainties described above are caused mainly by the measurement procedure or
availability of the physical properties. Off course the measurement devices self have also their
accuracy and the measured values fluctuate over time. Therefore it might be better to look at
the best and worst case per parameter which occur in the measurements. From this, one can
finally determine approximately a best and worst case for the measured pressure drop and
separation efficiency.

The dust concentration is calculated via the weight difference of the filter and the pumped
volume. For normalisation the oxygen content of the dry flue gas is also involved. For the
tubular filter device the pumped volume is calculated from the gas meter and the temperature
and pressure at the gas meter.
The weight is determined on a balance with an accuracy of 0.0001 g. Since the balance is
calibrated before each weight determination, it can be assumed that there is just a negligible

27
deviation from the real weight. Therefore the influence of the deviation in the weight
determination is neglected.
The readout on the gas meter goes at an accuracy of 0.0002 m3. According to Mawera this gas
meter is calibrated. The influence of the deviation in the pumped volume (which is in the order
of 1 m3) can therefore also be ignored.

The pressure and temperature at the Table 4.2: Characteristics of the measured parameters for
gas meter are manually read three the determination of the dust concentration. The
percentages are related to the average values
times (start, middle, end) during each
measurement. For each measurement
Absolute Temperature
the standard deviations can be pressure (in K)
calculated from these three values. The Standard deviation 15,1 mBar 0.24 oC
maximum positive and negative ( 2,28 % ) ( 0.08 % )
deviations in percent (from the average Minimum deviation 9,9 % 0,7 %
value) can also be determined for each ( lower limit )
measurement. The lower limit can be Maximum deviation 10,2 % 0,7 %
( upper limit )
taken as the minimum value of the
negative deviation from all measurements. The upper limit can be taken as the maximum value
of the positive deviation from all measurements. The real value shall be somewhere between
these two limits.
From these limits, the maximum positive and negative deviations in dust concentration can be
derived for the tubular filter device
For the STMG40 dust measuring device, the characteristics of the pumped volume are already
presented in section 3.2. However, they are determined with the gas meter from the tubular
filter device. This means that the characteristics from table 4.2 also influence the limits of the
measured pumped volume. Therefore a correction is needed with above parameters.
The minimum and maximum deviation limits for dust determination in front and after the
cyclone is given in table 4.3:

Tabel 4.3: Best and worse case in the deviation around the calculated dust concentrations. The deviation is
given in percent of the mean dust concentration. Note that this involves the dust concentration that is not
jet normalised to 13Vol.% O 2 .
STMG40 Tubular filter device
(in front of cyclone) (behind cyclone)
100 100 − 0.7 100 − 0.7
minimum deviation − 1 ⋅ 100% ≈ 16.0% − 1 ⋅100% ≈ 10.6%
(100 + 7.32) 100 + 10.2 100 + 10.2
100 100 + 0.7 100 + 0.7
maximum deviation − 1 ⋅100% ≈ 20.6% − 1 ⋅ 100% ≈ 11.8%
(100 − 7.32) 100 − 9.9 100 − 9.9

The limits from table 4.3 can be used to estimate the best and worst case in separation
efficiency measurements. This is allowed since the oxygen fraction in the flue gas duct changes
hardly, and can said to be equal before and after the cyclone. Therefore the fraction between
the concentrations in front and after the cyclone does not change after correction to 13 Vol.%
oxygen at dry flue gas. The area between the best and worst case defines the region in which it
is plausible to find the real values of the measured separation efficiency. The best case
(highest efficiency) is defined by the smallest dust concentration behind the cyclone, and the
largest dust concentration before of the cyclone. The worst case is defined in the opposite way.
The resulting two limits are expressed in table 4.4 on the next page.

28
Tabel 4.4: Best and worst case separation efficiency as follows from the measurement data.
Worst case: ⎧ 100 + 0.7 100 + 7.32 100 + 10.2 ⎫ C
⎛ ⎞⎛ ⎞⎛ ⎞ out C
η sep = 1 − ⎨⎜ ⎟⎜ ⎟⎜ ⎟⎬ ≈ 1 − 1.33 out
⎩⎝ 100 − 9.9 ⎠⎝ 100 ⎠⎝ 100 − 0 .7 C
⎠⎭ in Cin
Best case: ⎧⎛ 100 − 0.7 ⎞⎛ 100 − 7.32 ⎞⎛ 100 − 9.9 ⎞⎫ C out C
η sep = 1 − ⎨⎜ ⎟⎜ ⎟⎜ ⎟⎬ ≈ 1 − 0.75 out
⎩⎝ 100 + 10.2 ⎠⎝ 100 ⎠⎝ 100 + 0.7 ⎠⎭ C in Cin

Next to the dust concentrations, the temperatures and dynamic and static pressures in front and
after the cyclone are measured, in order to determine the flow rate in normal conditions and the
pressure drop over the cyclone.

The flow and pressure drop measurements are done automatically with a computer. For these
measurements it is not useful to look at the fluctuations around the measured value, since the
influence from the sensor-disturbance fluctuation is negligible compared to the influence from
fluctuations in the combustion unit. Therefore the average flow rate and pressure drop are
determined over 30 minutes (simultaneously with the dust measurements). One can now take a
look at the sensor accuracy instead of fluctuation characterisation, since information on sensor
accuracy is known for these sensors.

The temperature sensors are of the type Pt100 and are manufactured by ´Jumo´. According the
user manual [17], the sensors are of accuracy class B (norm: DIN EN 60 751). Class B gives an
accuracy of +(0.30 + 0.005T )oC. The average temperature before the cyclone is from the order
of 185 oC and after the cyclone in the order of 150 oC. This gives accuracies of T front + 1.23 oC
and T after + 1.05 oC.

The pressure sensors are manufactured by ´Huba Control´. They are pressure difference
sensors, type: 698. From the manufacturers website a technical datasheet is obtained from these
devices [18]. In this datasheet one can find the linearity and hysteresis error in percent from the
full scale. For the dynamic pressure, devices up to 5 mBar were used. For the static pressure
behind the cyclone a 50 mBar device was used and before the cyclone a 10 mBar device.
The resulting measurement errors for above devices are given in table 4.5.
Table 4.2: The deviations from the real values for the pressure loss
measurements.

Accuracy % Accuracy mBar


p stat , after ± (0.6% fs ) + 0.3 mBar
p stat , front ± (0.5% fs ) + 0.05 mBar
P dynamic ± (0.9% fs ) + 0.045 mBar

29
Next to the pressure sensors, also the influence of the position and direction of the Prandtl
tubes are involved in the flow rate determination. From the user manual [19] error estimation
can be done according the two graphs of figure 4.5.

Figure 4.5: The influence of the Prandtl tube direction on the error (left) and the error over the measured
air velocity at standard conditions. (right). The error (deviation) in percent is given on the vertical axis.
Left the rotation angle (direction) is given in degrees on the horizontal axis. Right the velocity is given on
the horizontal axis. ( Source: [19] )

The used Prandtl tubes had diameters of 2.5 mm and the flow fluctuated from 4 to 10 ms-1.
From figure 4.5 it can be concluded that the derived velocity (derived from the measured
dynamic pressure) contains a deviation of -0.5 to 0.4 percent from the real value, depending on
the fact that the angle between tube and flow direction is always smaller than 10 degrees.

From the determined deviations around the real value one can determine again a range around
the measured value, in which the real value is certainly expected. This range has limits at plus
and minus two times the deviation around the measured value, which can approximated as
given in table 4.4

Tabel 4.4: Best and worse case in the deviation around the normalized volumetric flow and the pressure
drop. The deviation is given in percent of the measured value. Note that this involves the volumetric flow
rate that is not jet normalised to 13Vol.% O 2 .
Minimum deviation Maximum deviation
12.7% 10.6%

The above accuracy analysis produces rectangles (uncertainty windows) around the
measurement points, in which the width and the height are determined by the above
approximated deviations of the volumetric flow rate and the separation efficiency or the
pressure drop. One can now say that the ‘real’ value of the measured point is certainly situated
in this window. The uncertainty windows are depicted in figures 4.6 and 4.7.

30
100

95

90

85
Rendement in [%]

80

75

model (PSD from fig A.2)


70
measurements

65
mean per number of cyclones

60
0 50 100 150 200 250 300 350 400 450 500
3
Flow in [m /h]

Figure 4.6: The uncertainty windows for the separation efficiency measurements. The
present single measurement points, the triangles are the mean over the measurements
on a certain number of single cyclones. The dashed boxes give the uncertainty window
for every measurement point.

45
Emperical relation eq.(4.1)

40 Measurement data

35

30
pressure drop in [mbar]

25

20

15

10

0
0 100 200 300 400 500 600
3
Flow in [m /h]
Figure 4.7: The uncertainty windows for the pressure drop measurements. The
dots are the measurement points. The dashed boxes give the uncertainty
window for every measurement point.

31
4.5 Geometrical parameter determination:

The characteristic particle size is a very important parameter in cyclone design and
performance, since a small characteristic particle size means a high performance. Therefore it
is useful to compare this parameter from the cyclone under investigation with the ones from
other cyclone designs. For such a comparison it is usual to rewrite the expression of the
characteristic particle size in a parameter group containing only geometric parameters and a
parameter group containing only physical parameters. The geometrical parameter group is then
called the geometrical parameter. This parameter is characteristic for a certain cyclone design,
and is expressed as X geo in equation (4.3). [12]. By comparing this geometrical parameter, one
can compare the performance of different cyclone designs, independent of their physical
environment.

μ G ⋅ Dc 3
d p ,50% = X geo⋅ (4.3)
Q ⋅ (ρ p − ρ g )

as the expression for the characteristic particle size was derived from force equilibrium, a
theoretical expression for the geometrical parameter can be obtained. It is however more
common to determine the geometrical parameter via the experimental determined ´real´
characteristic particle size.

From the experimental data, shown in figures 4.1 and 4.2 it is not possible to derive the ´real´
characteristic particle size directly. There are however two methods to determine the
geometrical parameter indirectly:

The first method is based on the measured separation efficiency, and is already short described
in the section above. If one knows the particle size distribution before the cyclone and the
fractional efficiency as a function of the particle size and characteristic particle size, than the
characteristic particle size can be determined by adjusting the characteristic particle size in the
fractional efficiency expression in such a way that the predicted separation efficiency is the
same as the measured one.
In the current case, neither the real particle size distribution nor the real fractional efficiency is
known. The ‘real’ characteristic particle size and corresponding value of the geometrical
parameter can only approximated when a particle size distribution and fractional efficiency
curve are assumed. This involves again the two major uncertainties which were described in
section 4.2.

The second method to determine the geometrical parameter is based on a theory from K.
Rietema. With some basic simplifications, he was able to derive a cyclone correlation number
which relates the total pressure difference over the cyclone to the characteristic particle size
[16]. Equation (4.4) shows this expression:

d p ,50% ⋅ (ρ p − ρ g ) (Δp )static + 12 ρ g ⋅ vi 2


2

Cy 50 = ⋅L⋅ (4.4)
μ ρg ⋅Q

32
Rietema stated that this cyclone number was only dependent on the geometry of a cyclone. He
also showed experimentally that there exists a minimum cyclone correlation number of 3.5 for
a maximum performance design. This design has ratios: L Dc = 5 , Do Dc = 0.34 ,
Din Dc = 0.28 , l Dc = 0.4 . Rietema presented the geometrical dependency of the cyclone
correlation number in two graphs. These graphs are given in figure 4.5.

Figure 4.5: The geometrical dependency of the cyclone correlation number. The right graph springs from
cyclones with a length over outer diameter ratio of 5. The left graph springs from cyclones with a fixed inlet
diameter over outer diameter ratio of 0.20 and a fixed exhaust diameter over outer diameter ratio of 0.04.
The red dot in the right graph presents the current cyclone under investigation. ( Source: [16] )

L D D
For the current cyclone under investigation the ratios hold: = 3.57 , o = 0.55 , in = 0.41 .
Dc Dc Dc
Since the inlet of the current cyclone is square, the inlet diameter is approximated by the
hydraulic diameter of the inlet: the ratio of four times the inlet surface over the length of the
inlet contour.
For these values, the position is shown in the right graph of figure 4.5. The reason that this dot
is out of limit is that the current cyclone uses a square spiral inlet (entrance on a larger diameter
than the outer cyclone diameter. Rietema used a round slot inlet for his experimental cyclones.
From the figure it can be seen that it is suggested that the contours of constant cyclone
correlation number are closed around the contour of Cy 50 = 3.5. If this is true a value of 6 to 8
is approximated for the cyclone correlation number of the cyclone under investigation. Since
the graph is representative for a L/D c -ratio of 5 and the current L/D c -ratio is smaller, the lower
value of 6 is taken for the cyclone correlation parameter. Equations 4.1 till 4.4 can be
combined with the cyclone number to express the geometrical parameter as a function of flow
rate and pressure drop:

2
2 ⋅ Cy 50 ⋅ Ai
X geo = (4.5)
Dc ⋅ L ⋅ (ζ + 1)
3

33
4.6 Cyclone design comparison:

Now if one knows the pressure drop curve or one knows the particle size distributions and the
fractional efficiency for different cyclone designs and sizes, one can compare their geometrical
parameters on the basis of section 4.4. This parameter is, as stated earlier, only dependent on
the design of the cyclone, and represents via the characteristic particle size in general the
fractional efficiency. So the smaller this geometrical parameter, the better is its separation
performance.
This comparison is shown in figure 4.6 for the current design, the optimum design from
Rietema and the cyclone design from Bradley. The last one has ratios: L Dc = 6.85 ,
Do Dc = 0.20 , Din Dc = 0.133 , l Dc = 0.33 , which corresponds with a cyclone number of
approximately 8, according figure 4.4. A pressure loss correlation for this design can be found
in [1].
In a first step, it is useful to look at the designs that have a same inlet diameter as the current
design, since that corresponds to same Reynolds numbers for different designs.
For the two above described design-types the geometric parameter curve is shown in figure 4.6.
In figure 4.6, also the experimental values of the geometrical parameter are shown; calculated
according the efficiency data (first method) and calculated according the pressure drop
correlation(second method, equation 4.1):

0.35
Efficiency data fit
Mawera`s cyclone
Rietema (1)
0.3 Bradley (1)

0.25
Geometric parameter Xgeo

0.2

0.15

0.1

0.05

0
0 50 100 150 200 250 300 350 400
3
Flow in [m /h]

Figure 4.6: The geometric parameter for different cyclone designs. The curves correspond to designs with
the same inlet area. The dots are determined via the fitting of the fractional efficiency (first method). The
lines are determined via the pressure loss correlations, using the Theory of Rietema (second method)

34
Table 4.2 gives the Tabel 4.2: Measurements corresponding to the designs of Rietema
1)
corresponding measures for the and Bradley, belonging to the curves in figure 4.6. Hydraulic
different cyclone designs diameter
corresponding to the curves in
figures 4.6, 4.7 and 4.8 Mawera’s Rietema Rietema Bradley Bradley
cyclone (1) (2) (1) (2)
From figure 4.6 it can be Dc 150 270.3 200 569.1 200
concluded that the Bradley Do 82.5 91.9 68 113.8 40
design is the one with the dc 80 49.2 56.0 617.5 217
optimum performance. However L 536 1351.7 1000 3898.5 1370
its dimensions are too large for l 176 108.1 800 187.8 66
1)
application in biomass d i 62.1 75.7 56 75.7 26.6
combustion units. Therefore it is possible to decrease the dimensions and volumetric flow rate
by increasing the number of cyclones in the multi-cyclone. The design trajectory is however
beyond the goal of this project, since the objective is to compare the different designs, to
confirm the possibility of improvement.

Optimal performance does not only mean maximal separation, but also the lowest operation
costs, which are controlled by the dissipated power in the cyclone. This dissipated power is the
product of the volumetric flow rate and the pressure loss over the cyclone.
So the optimal performance is defined as the best separation efficiency (smallest characteristic
particle size) possible at the lowest possible dissipated power.
Therefore drawback is made from geometric parameter to characteristic particle size again.
In figure 4.7 and 4.8 the pressure drop and characteristic particle size are given for the three
cyclone designs.

Figure 4.7: Characteristic particle size curves (left) and pressure drop curves (right) for the three different
designs as given in table 4.2.

35
4
10
Mawera`s cyclone
Rietema (1)
2
10 Rietema (2)
Bradley (1)
Bradley (2)
dissipated power by pressure loss [ W ]

0
10

-2
10

-4
10

-6
10

-8
10

-10
10

-12
10
0 50 100 150 200 250 300
characteristic particle size [ um ]

Figure 4.8: The Cyclones dissipated power versus characteristic particle size for the different designs from
table 4.2. The dotted lines give the cyclone designs with an outer diameter of 200 mm. The solid lines give
the designs with same inlet area. The dashed line gives the current cyclone.

The graphs from figure 4.7 can be combined to one graph of the dissipated power versus the
characteristic particle size. This graph is given in figure 4.8. From figure 4.8 and table 4.2 it
can be seen that the general trend is an improvement of costs versus performance for
decreasing cyclone dimensions and other designs. Therefore the suspicion of a possible
improvement of the separation efficiency at a decrease of pressure drop by changing the
cyclone geometry is confirmed.
If one goes a little further in the design process then one can find an improvement in
characteristic particle size and pressure drop (resulting in separation efficiency rise) at
reasonable dimensions. However one also needs to take in account the number cyclones needed
in the multi cyclone to satisfy these conditions at a total maximum flue gas flow of the plant.
From figure 4.7 and 4.8 and table 4.2 it can be seen every design is more efficient than the
current design.
So changing the design and dimensions of the cyclone can have a positive effect on dissipated
power and separation efficiency, if one is not limited by space, material and the number of
cyclones in the multi cyclone.

36
5. Recommendations on multi-cyclone geometry:

During the experiments it was noted that a lot of dust was settled on the bottom, walls and in
the corners in the entrance and exit room of the multi-cyclone. This is due turbulence and flow
rotations, caused by the rectangular design of the multi-cyclone entrance and exit room and the
positioning of the single-cyclones in the multi-cyclone.

One could however realise a reduction in dust settling by some small practical changes on the
current multi-cyclone design. A reduction in dust settling mostly means also a decrease in
turbulent flow and flow rotations in the multi cyclone entrance room. Those rotations and
turbulence consume power from the flow. So decreasing dust deposit should come together
with a reduction in dissipated power.

Some possible practical improvements are given


below:

1) Cyclones on one side of the multi-cyclone could


be replaced by cyclones from a mirrored type. In
this way the single-cyclone entrances are better
positioned behind the multi-cyclone inlet, resulting
in a more equal performance over all cyclones. The
disadvantage however is the rise of production
costs due to the requirement of two single-cyclone
designs instead of one in the current case.

Figure 5.1: An improvement in dust deposit


2) Also the supporting floor of the multi-cyclone by using mirrored cyclone geometries on one
entrance room can be adapted in a way that less side of the multi-cyclone
dust settling is possible. The single-cyclones are
now supported by the entrance room floor at the residue exit (bottom of single-cyclone). The
space between this supporting floor and the single-cyclone inlet acts as a deposition area for
dust. By lifting the supporting floor up to the bottom of the single-cyclone inlet, this deposition
area decreases enormously. This is shown in figure 5.2:

Figure 5.2: An improvement in dust deposit in the cyclone entrance room by raising the supporting floor up
to the single-cyclone inlets.

37
3) The inlet and exit duct of the multi cyclone are also quite small, compared to the multi-
cyclone entrance and exit rooms. Therefore, the flow needs to bend around a sharp edge to
reach the cyclones in the corners, which results in its turn in turbulence behind the edge. This
has quite an influence on the dissipated power. So by widening the multi-cyclone inlet and
outlet gradually to approximately the sizes of the cyclone entrance and exit room decreases this
effect. This design improvement is shown in figure 5.3:

Figure 5.3: An improvement in dust deposit by changing the width of the multi-cyclone inlet and outlet.

5) One could also increase the


efficiency of the inlet flow (and
decreasing dust deposit) by an
extension of the inlet ducts of the
single cyclones towards the inlet
of the multi-cyclone. In this way
one could realize in a quite simple
way a leak free inlet flow by
adapting the multi-cyclone
entrance towards the lengthened
single cyclone inlets.
This improvement also equalizes
the length the mean effective flow
path for all cyclones: This may be
explained as follows. The single-
cyclones in the first row have the Figure 5.4: An improvement in performance, pressure drop and dust
longest outlet tubes, but have deposit by extention of the single-cyclone inlet channels towards the
therefore a short length to the entrance of the multi-cyclone.
multi-cyclone inlet. The single-
cyclones in the last row have short outlet tubes (because of the stepwise supporting floor) but
have long a long length towards the multi-cyclone inlet. Equalized mean effective path means
also a more equal pressure drop over all single-cyclones, which result in a more equal
performance over all cyclones.
The main advantages of this improvement are: the reduction in size of the entrance room (the
room in which the inlets of the single cyclones are placed), the more equal performance over
all cyclones, a better and more equal pressure drop over all cyclones, and less dust deposit in
the multi-cyclone which result in a decreased need for cleaning. This idea is presented in figure
5.4:

38
6. Summary:
6.1 Conclusion:

From the preface, one can find that the goals of this essay were to investigate the separation
efficiency of the current multi-cyclone, to find an optimum in this the performance and to
investigate the possibility of performance improvement by changing the design.

Currently the multi cyclone is equipped with 10 cyclones at a flow rate of 140 normal cubic
meters per hour per cyclone. From the model and the measurements this corresponds with
separation efficiency of approximately 75 to 80 percent at a pressure drop of 3 to 4 millibar.
The theoretical model and measurements show that separation efficiencies of 90 to 95 percent
are reachable at realistic flow rates of 300 to 350 normal cubic meters per hour at a pressure
drop of approximately 12 to 17 millibar. So an improvement in separation efficiency is
possible. However from figures 4.1 and 4.2 it can be concluded that there exist no optimum for
separation efficiency. The separation efficiency approaches the 100 percent for higher flow
rates at a quadratic increase in pressure drop.
The overall combustion unit is designed for a multi-cyclone pressure drop of approximately 7
millibar which belongs to a flow rate of 230 to 240 normal cubic meters per hour and per
cyclone at a separation efficiency of approximately 85 to 90 percent. This means that the
current number of single cyclones is too much. For the design-requirement 6 single cyclones
are enough to digest the 1400 cubic normal meters per hour from the combustion unit.

From accuracy investigation it can overall be seen that the exactness of the measured and
predicted values are roughly within 20 to 25 percent for separation efficiency, within 10 to 12
percent for the flow rate measurements and 1 to 2 percent for the pressure drop prediction.

In the second part it can be found that a proper change in single-cyclone geometry and design
shall definitely result in a better performance at a smaller dissipated power. The overall trend
for the designs is a decrease in dissipated power at increased performance (smaller
characteristic particle size) for a decrease in single-cyclone dimensions.

From the total essay the general conclusion can be drawn that the current design can be
improved quit a lot by adaptation of the number of single cyclones, in combination with the
practical improvements given in chapter 5.
Further improvement and optimization is possible by adapting the single-cyclone geometry,
but will require a complete re-design of the multi-cyclone.

39
6.2 Recommendations on future investigation:
It is quite thinkable that the performance investigation on the current multi-cyclone is
continued in the future. Therefore some critical notes and recommendations for future
investigation are given below:

1) For an improvement of the theoretical prediction (model) one should gain more data on
separation efficiency, flow rate and pressure drop. It is further advised to do simultaneously
impactor measurements before and after the cyclone unit. In this way one gets insight in the
actual particle size distributions. From these particle size distributions one can derive the exact
fractional efficiency and finally better model for the ingoing particle size distribution. It is
recommended to do this also for different kind of biomass fuels.

2) For future measurements it is also advised to use equal measurement devises before and
after the cyclone in order to increase the accuracy of the measurements. This was not the case
in the current investigation. Also automatic dust measurements with a computer will increase
the accuracy of the separation efficiency measurements a lot.

2) Before a re-design is considered, some further investigation towards optimum single-


cyclone geometry is recommended. One should also pay attention on the prevention of dust
deposit, to guarantee long term performance.

3) There exist also a lot of other cyclone pressure drop reducing add-ons. These improvements
are not included in this report. It might be useful to gain some information about these pressure
drop reducing add-ons, before one starts a re-design.

40
7. References:
[1] Bart van Esch, Erik van Kemenade. (2004). Procestechnische constructies I.
TU-Eindhoven (NL). (4A460)

[2] VDI Verlag. (1996). Zyklonabscheider in der Energie- und Verfahrenstechnik.


Düsseldorf (DE):VDI Verlag GmbH. (ISBN: 3-18-091290-1)

[3] Prof. Dr.-Ing Wolfgang Fritz, Prof.Dipl.-Ing. Heinz Kern. (1990). Reinigung von
Abgasen. 2. Auflage. Würzburg (DE): Vogel Verlag und Druck KG.
(ISBN: 3-8023-0244-3)

[4] Daniel Wagner, Thomas Nussbaumer. (1994). Messverfahren zur Erfassung des
Emissionsverhaltens von Holzfeuerungen. Zürich (CH): Ingenieursbüro Verenum

[5] E. Weber, W. Brocke. (1973). Apparate und Verfahren der industriellen Gasreinigung
Band 1: Feststoffabscheidung. München (DE): R. Oldenbourg Verlag GmbH.

[6] Jianyi Chen, Mingxian Shi. (2006). A universal model to calculate cyclone pressure
drop. www.sciencedirect.com. Elsevier. Powder technology 171(2007) p188-p191.

[7] Excel file on Particle Size Distribution off coarse fly ash. (2005). Source: Joachim
Friesenbichler, Technical University Graz. The file contains the data from the PSD
from figure 2-2-2 on page 16 of the BIO-Aerosols final rapport: ´Aerosols in fixed-bed
biomass combustion – formation, growth, chemical composition, deposition,
precipitation and separation from flue gas´.

[8] Comité Européen de Normalisation {CEN}. (2004). Solid biofuels – Methods for the
determination of moisture content – Oven dry method – Part 2: Total moisture –
Simplified method. Brüssel (B): Management-center. Ref.Nr. CEN/TS 14774-2/2004D

[9] Verein Deutscher Ingenieure. (1993/1994). Manuelle Staubmessung in strömenden


Gasen . Gravimetrische Bestimmung der Staubbeladung. Filterkopfgeräte (4-12 m3/h).
Düsseldorf (DE). VDI Handbuch Reinhaltung der Luft, Band 4

[ 10 ] R. Zhang , P. Basu. (2004). A simple model for prediction of solid collection efficiency
of a gas-solid separator. www.sciencedirect.com. Elsevier. Powder technology
147(2004) p86-p93.

[ 11 ] Cristóbal Cortés , Antonia Gil. (2007). Modelling the gas and particle flow inside
cyclone separators. www.sciencedirect.com. Elsevier. Progress in Energy and
Combustion Science.

[ 12 ] Erik van Kemenade. Procestechnische constructies II.

[ 13 ] Excel file on biomass flue gas characteristics. (2005). Source: Joachim Friesenbichler,
Technical University Graz. The file contains data on aerosol, fly ash and fuel
characteristics. The data comes from pilot tests done at the Mawera test facility.

41
[ 14 ] Graz University of technology. (2003). BIO-Aerosols final rapport: ´Aerosols in fixed-
bed biomass combustion – formation, growth, chemical composition, deposition,
precipitation and separation from flue gas´.

[ 15 ] William C. Hinds. (1982). Aerosol Technology. 1. Edition. New-York (USA): Wiley-


Interscience.(ISBN: 0-471-08726-2)

[ 16 ] K. Rietema. (1961). Performance and design of hydrocyclones II & III.


Koninklijke/Shell-Laboratorium, Amsterdam. Chemical Engineering science 1961
Vol.15 p303-p319.

[ 17 ] Jumo. Typeblatt 90.2424: Wärmezahler-Wiederstandsthermometer mit Anslusskopf,


PTB zugelassen. Source: Mawera.

[ 18 ] Huba Control. 698 Pressure, vacuum and differential pressure module with or without
display 0 – 1/3/5/10/30/50 mbar. Source: www.hubacontrol.com

[ 19 ] User Manual Prandtl tubes: AIRFLOW Mikromanometer-Prüfsätze und AIRFLOW-


Staurohre. Source: Mawera

42
Appendices:

43
Symbolic declaration:(appendices)
A Required oxygen content for stoichiometric combustion [ mol ]
Ai Cyclone inlet surface [ m2 ]
C(d p ) Concentration distribution over particle size [ kgm-3 ]
dp Particle diameter [m]
d a.e. Aerodynamic particle diameter [m]
Dc Outer diameter of the cyclone separation chamber [m]
D0 Exhaust diameter of the cyclone separation chamber [m]
E Average absolute error [ … ] 1)
Fs Total contact area between the gas and the wall [ m2 ]
g Gravity constant [ ms-2 ]
K Constant [-]
L Length of the cyclone separation chamber [m]
m Mass [ kg ]
M… Molar mass [ kg mol-1 ]
n Swirl coefficient [-]
N Number of molecules [ mol ]
p pressure [ Pa ]
Q Volumetric flow [ m3s-1 ]
R Radius [m]
R Universal gas constant [ Jmol-1K-1 ]
Re t Reynolds number for tangential flow [-]
Re R Reynolds number for wall friction [-]
Rm R Specific gas constant [ Jkg-1K-1 ]
T Temperature [K]
u Moisture content of the dry biomass fuel [ mass% ]
v ax , 0 Average axial gas velocity [ ms-1 ]
vt Tangential gas velocity [ ms-1 ]
vt ,0 Tangential velocity at the position r = D o / 2 [ ms-1 ]
V Volume [ m3 ]
Vm Molar Volume [ m3mol-1 ]

εw Fraction of the dust load at the wall [-]


λw Wall friction coefficient [-]
λs Additional friction coefficient [-]
λ Friction coefficient [-]
μ mean value […]
μD Total dust load [ kg kg-1 ]
μG dynamic viscosity of the gas [ Pa s ]
ρG Gas density [ kgm-3 ]
ρp particle density [ kgm-3 ]
σ Standard deviation [ … ] 1)

[..] Volumetric percentage [ Vol% ]


1)
...Dependent on the unit of the involved parameter

44
A.1 Cyclone dimensions:

45
A.2 Dust measurement rapport:
Dust measurements on cyclone separation efficiency

Measurement number
Fuel
Moisture content mass% d.b. / mass% w.b.
Number of cyclones

Time & Date


Date dd:mm:yy

In front of cyclone Behind cyclone


start end start end
Time hh:mm :ss

Environmental values:
In front of cyclone Behind cyclone
start end start end
p environment mbar
o
T environment C

Dust measurement :
In front of cyclone Behind cyclone
start end start end
d probe inlet m
Bush number -
Bush weight g

Q-meter score m3 Fixed average rate at: 0.5092 nm3/h

p pump mbar
o
T pump C
C Oxygen , exhaust Vol.%

p dyn. exhaust Pa
p stat. exhaust Pa
Q exhaust m3/h
o
T exhaust C

Q pump , norm , 13 % Nm3/h


O2
Dust concentration Mg_

46
@ 13 Vol. % O 2 Nm3
A.3 Calculation of the exhaust gas density:

The density of the exhaust gases for different kinds of biomass fuels can be calculated in a
similar way as used in [4]. This calculation is based on the ideal gas law, assuming the
involved gases are ideal. To make the calculation easier, some other assumptions are made.
The general reaction between biomass and air that occurs in biomass combustion can be
depicted as:
u M fuel ⎛ 79 ⎞
CH m On + ⋅ H 2 O + λA ⋅ ⎜ O2 + N2 ⎟ →
100 M H 2O ⎝ 21 ⎠
⎛ ⎞
(1 − β ) ⋅ CO 2 + β ⋅ CO + ⎜⎜ m + u M fuel ⎟ ⋅ H 2 O + ⎛⎜ (λ − 1)A + β ⎞⎟ ⋅ O2 + λA 79 ⋅ N 2

⎝2 100 M H 2O ⎠ ⎝ 2⎠ 21

Herein air is assumed to be 20.4 Vol% O 2 and 79.6 Vol% N 2 . A presents the oxygen
requirement in [mol] for stoichiometric combustion as:

m n
A = 1+ − (A.1)
4 2

In a first approximation it can be assumed that biomass fuel compositions have all quite the
same composition as ordinary wood: m ≈ 1.44 and n ≈ 0.66 so that A ≈ 1.03 [mol]. It is
possible to change this in a future stage. Further it is assumed that the combustion is perfect, so
the factor β = 0 and soot and nitrogen oxide can be are neglected. Finally the fuel has a certain
moisture content u ( mass% d.b.). Determination of this moisture content is explained in
appendix A.4. Finally also the air humidity is neglected.
The main thing to do is now to calculate the specific gas constant R m . The relation between the
R

specific and universal gas constant can be derived from the ideal gas law:

p ⋅V R
N ⋅R = = m ⋅ Rm → Rm =
T m
N
( ) (A.2)

In above expressions R and R m are respectively the universal and specific gas constant in
R

[Jmol-1K-1] and [Jkg-1K-1], V is a volume in [m3], N is the number of molecules in [mol], m is


the mass in [kg] and p is the pressure in [Pa].

So the fraction between the total exhaust gas mass m and the total molecules in the gas N gives
the relation between the universal and specific case. Since it is generally known that the molar
volume V m ( given in [m3mol-1] ) is approximately constant for all gases, the volumetric
fractions are also the same as the molar fractions. Fraction m/N in equation (A.2) can now be
written as:

N CO2 M CO2 + N H 2O M H 2O + N O2 M O2 + N N 2 M N 2
=
⎛m⎞ N CO2 + N H 2O + N O2 + N N 2
⎜ ⎟ (A.3)
⎝N⎠ [CO2 ]⋅ M CO + [H 2 O] ⋅ M H O + [O2 ]⋅ M O + [N 2 ] ⋅ M N
= 2 2 2 2

100

47
Herein the brackets [..] stands for the real volumetric fraction in [Vol%] and M … for the molar
mass in [kgmol-1].
Now there are several possibilities. If the volumetric fractions can be measured for CO 2 and
H 2 O, the real specific gas constant can be calculated according (A.4)

8.3145 ⋅ 10 2 (A.4)
Rm =
⎛ 100 − [CO 2 ] − [H 2 O ] ⎞
[CO 2 ] ⋅ M CO + [H 2 O ] ⋅ M H 2O +⎜
100
(
⎟ ⋅ 20.4 M O2 + 79.6 M N 2 )
⎝ ⎠
2

In our case we prescribe the air excess ratio λ and the moisture content u of the fuel. The
fractions can than be deduced from the assumed chemical reaction at the previous page:

⎛m u M Fuel ⎞⎟ ⎛ 79.6 ⎞
1+ ⎜ + + ((λ − 1)A) + ⎜ λA ⎟
⎜ 2 100 M H O ⎟ ⎝ 20.4 ⎠
Rm = 8.3145 ⋅ ⎝ 2 ⎠ (A.5)
⎛m u M Fuel ⎞⎟ ⎛ 79.6 ⎞
M CO2 +⎜ + M + ((λ − 1)A)M O2 + ⎜ λA ⎟M N 2
⎜ 2 100 M H O ⎟ H 2O ⎝ 20.4 ⎠
⎝ 2 ⎠

According the ideal gas law the gas density can now be written as:

p
ρG = (A.6)
Rm ⋅ T

This calculation can now be implemented in the matlab model.

A.4 Normalized conditions:

In order to compare the different measurements with each other, they have to be normalized to
a specific condition. The normal condition: This means that volumetric values are recalculated
with the universal gas law to normal conditions:

Tn = 273.14 Kelvin
pn = 1013 millibar

Biomass fuels have quite high moisture contents which can vary a lot between two fuels or
even two batches of the same fuel. This means the volumetric fraction of water vapour in the
flue gases varies which influences the actual concentrations of dust and other species.
Therefore reference is made to the dry flue gas containing no water vapour. Since most
measurement devices are sensitive to moisture, the flue gas used in experimental equipment is
usually dried. So measured volumes are then automatically considered dry.

Biomass combustion happens always under conditions of air excess. This air excess varies over
time and position and will be different for different fuel loads. When the air excess increases
concentrations will decrease. This means that the dust concentration varies when the air excess
changes. Therefore reference is made to a certain air excess:

48
This excess is expressed as a fixed oxygen concentration in the flue gas. It is commonly to use
a dry flue gas oxygen concentration of 13 % as a reference value for the excess air amount.

With this, the total dry flue gas volume(rate) can be rescaled to 13 % oxygen, using the
reaction equation from appendix A.2, expressing λ as oxygen available divided by oxygen
required and assuming perfect combustion because of the air excess:

21 ⎞

(1 − A) + 100 ⎜⎜ ⎟A
V[ O2 ] 21 ⎝ 21 − [O2 ] ⎟⎠
= (A.7)
V[13%]
(1 − A) + 100 ⎛⎜ 21 ⎞⎟ A
21 ⎝ 21 − 13 ⎠

In appendix A.2 it is stated that the stoichiometric oxygen requirement A is approximately


1.03. If this is filled into equation (A.7) one finds:

⎛ 103 ⎞
− 0.03 + ⎜⎜ ⎟
⎛ V[O2 ] ⎞ ⎝ 21 − [O2 ] ⎟⎠ 21 − 13
⎜ ⎟= ≈ (A.8)
⎜V ⎟ ⎛ 103 ⎞ 21 − [O2 ]
⎝ [13%] ⎠ − 0.03 + ⎜ ⎟
⎝ 21 − 13 ⎠

This last expression is commonly used in most literature

For example taking the dust concentration in mg nm-3:

mdust mdust ⎛ V[ O ] ⎞
C dust ,norm ,13% = = ⋅⎜ 2 ⎟
Vnorm ,13% Vnorm ,[ o2 ] ⎜V ⎟
⎝ [13%] ⎠

To complete the summarization of the normalized conditions: The Aerodynamic particle


diameter according Hinds Aerosol technology [15]:

Related to the particle diameter via:

VTS (ρ p , d p ) = VTS (ρ = 1000, d a.e ) ρ p ⋅ d p 2 = 1000 ⋅ d a.e 2

49
A.5 Moisture content:

To calculate the flue gas density and to refer to dry flue gases in experiments, it is important to
know the moisture content of the fuel. This moisture content is determined according to the
CEN guideline CEN/TS 14774-2, [8].
This guideline gives the simplified method for determination of the moisture content, using a
drying furnace and fuel samples. They refer to the moisture content of the wet basis.
At Mawera these fuel samples are stored in visually clean and dried one litre canisters.

First the empty dry and clean canister is weighted. The canister is then filled with at least 300 g
of fuel, and is weighted directly. After weight determination, the sample is stored in a drying
furnace at a temperature of 105 (+/- 2) degrees Celsius. The storing time depends on the kind
of fuel and the amount of moisture. Normally a storing time of 24 hours should be satisfactory
to gain constant dry weight.
Since biofuels are very hygroscopic, they are weighted within 15 seconds after they are taken
out of the drying furnace.

The moisture content can now be calculated according equation (A.9):

mwet − mdried
u= ⋅ 100% (w.b.) (A.9)
mwet − mcanister

50
A.6 Determination of the friction coefficient:
For the calculation of the characteristic particle size and the characteristic dust load factor in
chapter 2, a friction coefficient is needed. This friction coefficient depends on many factors,
such as the dust load, wall friction and friction in the gas. In the book of ‘Apparate und
Verfahren der industriellen gasreinigung’[5] a method is given to determine this friction
coefficient. Herein the friction coefficient contains a contribution for the wall friction and a
contribution for the friction caused by dust flows at the wall:

ρg v ax ,0 −5
λ = λw + λs ⋅ μ D ⋅ ε w ⋅ ⋅ ⋅ Re t 8 (A.7)
ρ p Do ⋅ g

λ w presents the wall friction coefficient, λ s


an additional friction coefficient, μ D is the
dust loading factor in [ kg kg-3 ], ε w is the
fraction of the dust load at the wall, ρ G and
ρ p the gas and particle density in [ kgm-3 ],
v ax , 0 the average axial velocity in [ ms-1 ]
which can be calculated from the volumetric
flow and the exhaust duct cross section. D 0
is the exhaust diameter of the cyclone
separation chamber in [ m ], g is the gravity
constant in [ ms-2 ] and Re t is the tangential
Reynolds number at the exhaust radius.
As can be seen, the additional part is
dependent on many parameters, such as
flow conditions, densities, dust loading and
separation efficiencies. According [3] and
[5] equation (A.7) can be approximated with
averaged values to:
figure A.1: The wall friction as function of the
(
λ ≈ λw ⋅ 1 + K ⋅ μ D ) (A.8)
Reynoldsnumbers and different relative wall
roughness. Source: ‘Apparate und Verfahren in der
industriellen Gasreinigung’
Herein the most used value for K is 3 and with very low dust loadings 2.
The wall friction coefficient is also dependent on a Reynolds number, and can be roughly
determined using figure A.1. Herein the wall friction coefficient between a gas and a steel
cyclone wall is given as function of a Reynolds number for different wall roughness.
The Reynolds number for this figure is approximately given as:

Do ⋅ v t , 0 ⋅ ρ g
2
Q Dc
Re R = vt , 0 ≈ (A.9) (A.10)
⎛D ⎞ Ai Do
μ g ⋅ 4 ⋅ L ⋅ ⎜⎜ c − 1⎟⎟
⎝ Do ⎠

51
The tangential velocity can be approximated with the free vortex relation as is shown in
equation (A.10). In above expressions μ G presents the dynamic viscosity of the gas in [ Pa s ],
L the length of the cyclone separation chamber in [ m ], D c the outer diameter of the cyclone
separation chamber in [ m ], Q the volumetric flow in [ m3s-1 ] and A i the cyclone inlet surface
in [ m2 ].
If we now assume a very smooth cyclone wall and an average inlet flow of about 200 cubic
meters per hour, the value of the mentioned Reynolds number Re R is about: 3 ⋅ 10 3 . According
R

figure A.1 this corresponds with a wall friction coefficient of about 0.005. For higher flows, the
mentioned Reynolds number increases, but as can be seen from figure A.1, this has almost no
influence on the wall friction coefficient.

A.7 Particle size distribution:

To make an overview of the 250


Bark
separation efficiency, the
distribution of the dust Spruce

concentration over the particle 200 Waste wood


particle concentration in [ mg/Nm ]
3

size is needed. Mostly these Fibre Board

kinds of characteristics are


given as the particle 150

concentration in mass per


normal volumetric unit over 100
the aerodynamic particle
diameter. These kinds of
characteristics are often 50

measured with an impactor. An


impactor stores particles,
isokenetically extracted from a 0
-1 0 1 2 3
10 10 10 10 10
gas flow, on different shells particle size in [ um ]

according particle size areas.


The weight of these different Figure A.2: The particle size distribution of coarse fly ash for 4
shells, combined with the different fuels. The interrupted lines give log-normal distributions as3
extracted volume, determines possible approximations. The distribution is corrected to 300 mg/nm
such a particle size distribution.
According investigation towards biomass flue gases, [14], a bimodal particle density
distribution holds for the flue gases This distribution contains a small peak below one micron
(aerosols) and a peak around 30 micrometers (course fly ash). Since it is assumed particles
below one micrometer pass the filters from the dust measurement device anyway, the first peak
can be neglected.

For the model, distributional data on coarse fly ash (> 1 µm) in biomass combustion gases is
gained from the University of Graz [7]. This data contains the PSDs of four different fuels.
The file contains distributional data from biomass combustion gases, right after the combustion
chamber. The distributions from the data are shown in figure A.2:

52
As can be seen from figure A.2, the distributions can be approximated with a distribution over
the logarithm of the particle size, a log-normal distribution. Equation (A.11) expresses this
distribution function:

⎛ ⎛ log(d p ) − μ ⎞ 2 ⎞
C (d p ) =
1
⋅ exp⎜ − ⎜⎜ ⎟⎟ ⎟ (A.11)
σ ⋅ 2π ⎜ 2σ ⎠ ⎟⎠
⎝ ⎝

The corresponding mean particle size and the standard deviations are given in table A.1:

Tabel A.1: Values corresponding to the log-normal distribution as an approximation for the particle size
distribution. 1) The standard deviation and the mean particle size in the log-normal distribution are the
logarithms of the particle sizes in the table.
Fuel Mean particle size 1) Standard deviation 1)
Bark 22,4798 1,7942
spruce 44,9020 2,1288
waste wood 44,9020 1,7139
Fibre board 44,9020 2,0194

For use in a matlab model, this data is interpolated over its corresponding particle size, since
this gives better results for the distribution than a log-normal distribution.
In order to determine how many steps the interpolation requires, the average absolute error
between the real and interpolated points is calculated. This average absolute error is given as:

∑E 2 0.35
i Bark
1 Spruce
E= ⋅ i
⋅ 100% (A.12) Waste wood
absolute average error: Error / sum(C(dp)) [ % ]

0.3
UD i Fibre Board

0.25

Herein is E i the difference between the


value at a specific particle size and the 0.2

corresponding value resulting from the


interpolation at this particle size point, 0.15

in [mg/nm3]. U D is the total dust load of 0.1

the gas in [mg/nm3] and E is the


average absolute error in percent. 0.05

The result is shown in Figure A.3. From


approximately 200 steps and more, the 0
50 100 150 200 250 300 350 400

average absolute error does not decrease number of steps in particle size range

anymore, so 200 steps should satisfy for Figure A.3: Average absolute error, caused by the interpolation
of the PSD-data. The error between the data and the interpolated
the interpolation of the PSD.
function goes to a constant value for 200 steps or more

For the model it is important that the total dust concentration, so the sum of the particle mass
concentration over the particle size, equals the total dust concentration in front of the cyclone
in the test facility. This total dust concentration is about 300 [mg/nm3]. After this correction it
can be seen from the distributions in figure A.2, that with the exceptions of ´bark´ and ´waste
wood´, the distributions are approximately the same. Therefore spruce is taken as the
distribution which presents the wood chips, used at the experiments.

53
A.8 Determination of the total cyclone wall-gas contact surface area:

For the calculation of the friction between the cyclone walls and the gas, in order to determine
the pressure drop, the total contact area between gas and cyclone wall is needed.
This area can be described by the geometry of the cyclone. In the case of a single cyclone, also
the residue bunker is included, because the swirling core flow might also introduce a swirl in
the bunker. In the case of a multi-cyclone, this effect is neglected because the several single
cyclones have the same bunker, where multiple swirls damp each other.

The different cyclone components which participate to the contact surface are:

- top cover Atop =


π
4
(D c
2
− Do
2
) ( A.11 )
- exhaust cylinder Aexh = πDo l ( A.12 )
- Outer cylinder (wall) Acyl = πDc Lcyl ( A.13 )

π (Dc − d c )2
- Conical part Acon = (Dc − d c ) 2
Lcon + ( A.14 )
2 4
- Dust bunker (eventually)

In the above expressions L cyl and L con are the lengths from the cylindrical and conical parts of
the cyclone in [m]. d c is the residue exhaust diameter in [m].
The total gas – wall contact surface area can obviously be calculated from the summation:

Fs = Atop + Aexh + Acyl + Acon ( A.15 )

In combination with appendix A.1 this gives:

Fs ≈ 234974.861 mm2

54
A.9 Swirl coefficient:

As stated in the chapter about cyclone theory, the free vortex distribution of the tangential
velocity annular region of cyclone the separation chamber is not a perfect free vortex, but a
reduced free vortex. This reduce is introduced via an exponent, the swirl exponent n.
This relation can be depicted as follows:

vt ⋅ r n = constant instead of: vt ⋅ r = constant

For the general flow is a cyclone, n varies from one at the outer part of the flow in a cyclone, to
minus one at the center flow of the cyclone (solid body rotation) [11].
For the general outer flow, n varies approximately between 0.4 and 0.8, and is very dependent
on geometrical and physical properties. Therefore n is mostly derived empirically from
experimental results.

In [7] an empirical relation is given for this swirl exponent:

⎧⎪ ⎛ l − Hi ⎞
−0.5
⎫⎪ 4 ρ g Ai vi
n = 1 − exp ⎨− 0.26 Re 0.12 ⎜⎜1 + ⎟
⎟ ⎬ Re = ( A.16 )
⎪⎩ ⎝ Bi ⎠ ⎪⎭ πDo μ g

Herein the swirl exponent is correlated to the Reynolds number and the cyclone inlet geometry.

Another equation to approximate the swirl coefficient in the outer flow of the cyclone is given
in [11], according ´Alexander RMCK´:

( )⋅ ⎛⎜ 283
0.3
0.14 T ⎞
n = 1 − 1 − 0.67 Dc ⎟ (A.17)
⎝ ⎠

In this last expression, T presents the gas temperature in [K].

55

You might also like